|
exchange
newsgroups
|
|||||||||||||||||||||||
|
|||||||||||||||||||||||
Suggestions wanted for drive configuration for new Exchange 2003 serverExchange 2003 system. We're in the process of building a new Exchange 2003 server as a replacement for an existing server. It's an HP ML570 G2 with four 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but it's newer than the original one. Also, the original one has gone through several Windows and Exchange service packs, hot fixes, etc., and I'd like to get a clean 2003 R2 O/S install. I'm going to build the new server then migrate the mailboxes then decommission the old one. Here's the situation: The existing server holds about 2,500 mailboxes. This is a mailbox server only, OWA and the internet connectivity is handled by other servers. It is currently set up as two storage groups, with one database each. Each database (with STM file) is about 90GB. The current arrangement has each database on its own drive RAID5 array and each of logs on its own 2 drive RAID1. The replacement server has 14 drive bays. Two will have the O/S (RAID1), so that leaves 12 for Exchange. I have available a mix of 36, 72, and 146GB 15K U320 SCSI drives. My preferred arrangement is a single storage group/single database on a 10 drive RAID1 for the data and a 2 drive RAID1 for the logs. However, I'd like to keep the database size under 100G, so that means two databases. This brings up several options: Should I go with one storage group with two info stores, or two SGs with one database each? I seem to remember when we set these up the recommended configuration for multiple databases was multiple storage groups, but is that still the recommended solution? With multiple databases, should I have each database on its own RAID1 array? I could do something like two 4 drive RAID1 arrays (one for each database), then two 2 drive RAID1 for the logs, one for each SG (if I use two SG's). If I created one big drive array and put both databases on it, would the performance of having a single larger array with more spindles make up for putting two databases on the same array? With two storage groups, what about both databases on one 8 drive RAID1 array, and two separate two drive RAID1 arrays, one for each set of logs? I know there's all kinds of variations on this, but I'm trying to get the best options on limited hardware. Hopefully this won't be a long term need; we've submitted in the 2008 budget for new hardware & migration to Exchange 2007 (we put it in this year but the hardware budget was cut). Any suggestions would be appreciated. Mike O. With the number of drives you have, I would try one storage group. Two
drives RAID-1 for the OS, two drives RAID-1 for the transction logs, and 12 drives in a RAID-1+0 for the databases, but the feasibility of that depends on the inventory of drives you have. -- Show quoteEd Crowley MVP - Exchange "Protecting the world from PSTs and brick backups!" "Mike O" <put_the_spam@the.can> wrote in message news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... > I'm looking for some suggestions regarding hard drive configuration for an > Exchange 2003 system. > > We're in the process of building a new Exchange 2003 server as a > replacement for an existing server. It's an HP ML570 G2 with four 3.0 > Ghz processors, 4GB of RAM. It's not new hardware, but it's newer than > the original one. Also, the original one has gone through several Windows > and Exchange service packs, hot fixes, etc., and I'd like to get a clean > 2003 R2 O/S install. I'm going to build the new server then migrate the > mailboxes then decommission the old one. > > Here's the situation: The existing server holds about 2,500 mailboxes. > This is a mailbox server only, OWA and the internet connectivity is > handled by other servers. It is currently set up as two storage groups, > with one database each. Each database (with STM file) is about 90GB. The > current arrangement has each database on its own drive RAID5 array and > each of logs on its own 2 drive RAID1. > The replacement server has 14 drive bays. Two will have the O/S > (RAID1), so that leaves 12 for Exchange. I have available a mix of 36, > 72, and 146GB 15K U320 SCSI drives. My preferred arrangement is a single > storage group/single database on a 10 drive RAID1 for the data and a 2 > drive RAID1 for the logs. However, I'd like to keep the database size > under 100G, so that means two databases. > > This brings up several options: > > Should I go with one storage group with two info stores, or two SGs with > one database each? I seem to remember when we set these up the > recommended configuration for multiple databases was multiple storage > groups, but is that still the recommended solution? > > With multiple databases, should I have each database on its own RAID1 > array? I could do something like two 4 drive RAID1 arrays (one for each > database), then two 2 drive RAID1 for the logs, one for each SG (if I use > two SG's). > > If I created one big drive array and put both databases on it, would the > performance of having a single larger array with more spindles make up for > putting two databases on the same array? > > With two storage groups, what about both databases on one 8 drive RAID1 > array, and two separate two drive RAID1 arrays, one for each set of logs? > > I know there's all kinds of variations on this, but I'm trying to get the > best options on limited hardware. Hopefully this won't be a long term > need; we've submitted in the 2008 budget for new hardware & migration to > Exchange 2007 (we put it in this year but the hardware budget was cut). > > Any suggestions would be appreciated. > > Mike O. > "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message With one storage group, but still use 2 databases? I know the Exchange can news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... > With the number of drives you have, I would try one storage group. Two > drives RAID-1 for the OS, two drives RAID-1 for the transction logs, and > 12 drives in a RAID-1+0 for the databases, but the feasibility of that > depends on the inventory of drives you have. handle big databases, but the BPA flags it if the database is over 100GB. Besides, I really don't ever want to do an emergency or mailbox restore on a 200GB database.. Show quote > -- > Ed Crowley > MVP - Exchange > "Protecting the world from PSTs and brick backups!" > > "Mike O" <put_the_spam@the.can> wrote in message > news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >> I'm looking for some suggestions regarding hard drive configuration for >> an Exchange 2003 system. >> >> We're in the process of building a new Exchange 2003 server as a >> replacement for an existing server. It's an HP ML570 G2 with four 3.0 >> Ghz processors, 4GB of RAM. It's not new hardware, but it's newer than >> the original one. Also, the original one has gone through several Windows >> and Exchange service packs, hot fixes, etc., and I'd like to get a clean >> 2003 R2 O/S install. I'm going to build the new server then migrate the >> mailboxes then decommission the old one. >> >> Here's the situation: The existing server holds about 2,500 mailboxes. >> This is a mailbox server only, OWA and the internet connectivity is >> handled by other servers. It is currently set up as two storage groups, >> with one database each. Each database (with STM file) is about 90GB. >> The current arrangement has each database on its own drive RAID5 array >> and each of logs on its own 2 drive RAID1. >> The replacement server has 14 drive bays. Two will have the O/S >> (RAID1), so that leaves 12 for Exchange. I have available a mix of 36, >> 72, and 146GB 15K U320 SCSI drives. My preferred arrangement is a >> single storage group/single database on a 10 drive RAID1 for the data and >> a 2 drive RAID1 for the logs. However, I'd like to keep the database >> size under 100G, so that means two databases. >> >> This brings up several options: >> >> Should I go with one storage group with two info stores, or two SGs with >> one database each? I seem to remember when we set these up the >> recommended configuration for multiple databases was multiple storage >> groups, but is that still the recommended solution? >> >> With multiple databases, should I have each database on its own RAID1 >> array? I could do something like two 4 drive RAID1 arrays (one for each >> database), then two 2 drive RAID1 for the logs, one for each SG (if I use >> two SG's). >> >> If I created one big drive array and put both databases on it, would the >> performance of having a single larger array with more spindles make up >> for putting two databases on the same array? >> >> With two storage groups, what about both databases on one 8 drive RAID1 >> array, and two separate two drive RAID1 arrays, one for each set of >> logs? >> >> I know there's all kinds of variations on this, but I'm trying to get the >> best options on limited hardware. Hopefully this won't be a long term >> need; we've submitted in the 2008 budget for new hardware & migration to >> Exchange 2007 (we put it in this year but the hardware budget was cut). >> >> Any suggestions would be appreciated. >> >> Mike O. >> > > Exactly. The reason you will want to use two databases is in the event one
fails your time to restore is half that of a 100 GB database. You could break it down into four databases if you want. -- Show quoteEd Crowley MVP - Exchange "Protecting the world from PSTs and brick backups!" "Mike O" <put_the_spam@the.can> wrote in message news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... > > "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message > news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >> With the number of drives you have, I would try one storage group. Two >> drives RAID-1 for the OS, two drives RAID-1 for the transction logs, and >> 12 drives in a RAID-1+0 for the databases, but the feasibility of that >> depends on the inventory of drives you have. > > With one storage group, but still use 2 databases? I know the Exchange > can handle big databases, but the BPA flags it if the database is over > 100GB. Besides, I really don't ever want to do an emergency or mailbox > restore on a 200GB database.. > > >> -- >> Ed Crowley >> MVP - Exchange >> "Protecting the world from PSTs and brick backups!" >> >> "Mike O" <put_the_spam@the.can> wrote in message >> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>> I'm looking for some suggestions regarding hard drive configuration for >>> an Exchange 2003 system. >>> >>> We're in the process of building a new Exchange 2003 server as a >>> replacement for an existing server. It's an HP ML570 G2 with four 3.0 >>> Ghz processors, 4GB of RAM. It's not new hardware, but it's newer than >>> the original one. Also, the original one has gone through several >>> Windows and Exchange service packs, hot fixes, etc., and I'd like to get >>> a clean 2003 R2 O/S install. I'm going to build the new server then >>> migrate the mailboxes then decommission the old one. >>> >>> Here's the situation: The existing server holds about 2,500 mailboxes. >>> This is a mailbox server only, OWA and the internet connectivity is >>> handled by other servers. It is currently set up as two storage groups, >>> with one database each. Each database (with STM file) is about 90GB. >>> The current arrangement has each database on its own drive RAID5 array >>> and each of logs on its own 2 drive RAID1. >>> The replacement server has 14 drive bays. Two will have the O/S >>> (RAID1), so that leaves 12 for Exchange. I have available a mix of 36, >>> 72, and 146GB 15K U320 SCSI drives. My preferred arrangement is a >>> single storage group/single database on a 10 drive RAID1 for the data >>> and a 2 drive RAID1 for the logs. However, I'd like to keep the >>> database size under 100G, so that means two databases. >>> >>> This brings up several options: >>> >>> Should I go with one storage group with two info stores, or two SGs with >>> one database each? I seem to remember when we set these up the >>> recommended configuration for multiple databases was multiple storage >>> groups, but is that still the recommended solution? >>> >>> With multiple databases, should I have each database on its own RAID1 >>> array? I could do something like two 4 drive RAID1 arrays (one for each >>> database), then two 2 drive RAID1 for the logs, one for each SG (if I >>> use two SG's). >>> >>> If I created one big drive array and put both databases on it, would the >>> performance of having a single larger array with more spindles make up >>> for putting two databases on the same array? >>> >>> With two storage groups, what about both databases on one 8 drive RAID1 >>> array, and two separate two drive RAID1 arrays, one for each set of >>> logs? >>> >>> I know there's all kinds of variations on this, but I'm trying to get >>> the best options on limited hardware. Hopefully this won't be a long >>> term need; we've submitted in the 2008 budget for new hardware & >>> migration to Exchange 2007 (we put it in this year but the hardware >>> budget was cut). >>> >>> Any suggestions would be appreciated. >>> >>> Mike O. >>> >> >> > Thanks for the info. I figured that having one disk array with more
spindles would give the best disk I/O, but I was concerned that multiple databases on the same set of disks would cause performance issues and offset better I/O of a single array. Mike O. Show quote "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message news:%23GQJDgI$HHA.748@TK2MSFTNGP04.phx.gbl... > Exactly. The reason you will want to use two databases is in the event > one fails your time to restore is half that of a 100 GB database. You > could break it down into four databases if you want. > -- > Ed Crowley > MVP - Exchange > "Protecting the world from PSTs and brick backups!" > > "Mike O" <put_the_spam@the.can> wrote in message > news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... >> >> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >> news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >>> With the number of drives you have, I would try one storage group. Two >>> drives RAID-1 for the OS, two drives RAID-1 for the transction logs, and >>> 12 drives in a RAID-1+0 for the databases, but the feasibility of that >>> depends on the inventory of drives you have. >> >> With one storage group, but still use 2 databases? I know the Exchange >> can handle big databases, but the BPA flags it if the database is over >> 100GB. Besides, I really don't ever want to do an emergency or mailbox >> restore on a 200GB database.. >> >> >>> -- >>> Ed Crowley >>> MVP - Exchange >>> "Protecting the world from PSTs and brick backups!" >>> >>> "Mike O" <put_the_spam@the.can> wrote in message >>> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>>> I'm looking for some suggestions regarding hard drive configuration for >>>> an Exchange 2003 system. >>>> >>>> We're in the process of building a new Exchange 2003 server as a >>>> replacement for an existing server. It's an HP ML570 G2 with four >>>> 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but it's newer >>>> than the original one. Also, the original one has gone through several >>>> Windows and Exchange service packs, hot fixes, etc., and I'd like to >>>> get a clean 2003 R2 O/S install. I'm going to build the new server >>>> then migrate the mailboxes then decommission the old one. >>>> >>>> Here's the situation: The existing server holds about 2,500 mailboxes. >>>> This is a mailbox server only, OWA and the internet connectivity is >>>> handled by other servers. It is currently set up as two storage >>>> groups, with one database each. Each database (with STM file) is about >>>> 90GB. The current arrangement has each database on its own drive RAID5 >>>> array and each of logs on its own 2 drive RAID1. >>>> The replacement server has 14 drive bays. Two will have the O/S >>>> (RAID1), so that leaves 12 for Exchange. I have available a mix of 36, >>>> 72, and 146GB 15K U320 SCSI drives. My preferred arrangement is a >>>> single storage group/single database on a 10 drive RAID1 for the data >>>> and a 2 drive RAID1 for the logs. However, I'd like to keep the >>>> database size under 100G, so that means two databases. >>>> >>>> This brings up several options: >>>> >>>> Should I go with one storage group with two info stores, or two SGs >>>> with one database each? I seem to remember when we set these up the >>>> recommended configuration for multiple databases was multiple storage >>>> groups, but is that still the recommended solution? >>>> >>>> With multiple databases, should I have each database on its own RAID1 >>>> array? I could do something like two 4 drive RAID1 arrays (one for each >>>> database), then two 2 drive RAID1 for the logs, one for each SG (if I >>>> use two SG's). >>>> >>>> If I created one big drive array and put both databases on it, would >>>> the performance of having a single larger array with more spindles make >>>> up for putting two databases on the same array? >>>> >>>> With two storage groups, what about both databases on one 8 drive RAID1 >>>> array, and two separate two drive RAID1 arrays, one for each set of >>>> logs? >>>> >>>> I know there's all kinds of variations on this, but I'm trying to get >>>> the best options on limited hardware. Hopefully this won't be a long >>>> term need; we've submitted in the 2008 budget for new hardware & >>>> migration to Exchange 2007 (we put it in this year but the hardware >>>> budget was cut). >>>> >>>> Any suggestions would be appreciated. >>>> >>>> Mike O. >>>> >>> >>> >> > > The IO pattern of exchange databases is random. If you put two databases on
the same set of spindles, it's still random. The only difference is an increse in the load the spindles must support. The IO pattern for logs is sequential. If you put two sets of logs on the same set of physical spindles, the pattern changes to one that is less sequential and there are performance reprecussions. Stores within a storage group share the same set of log files. Each storage group has a seperate set of logs. The current MS guidance for Storage Group design is to start with the largest number, even if you only have one store per store group. This minimizes contention for a given set of log files. Log writes are the most user visible aspect of Exchange IO. http://support.microsoft.com/kb/890699/en-us Let's start by considering only best practices without regard to the number of spindles you have available. To provide fault tolerance, the OS volume should be mirrored. Each set of log files should be on physically seperate spindles. RAID 1. 10, or 0+1 is preferred for databases. RAID 5 does not provide an optimal (or arguably even acceptable) tradeoff between fault tolerance and performance. If you go with two storage groups, then you have 3 RAID 1 mirrors (one for the os and one for each set of logs). For the databases you could use 1 large RAID 10, or two smaller RAID 10s. From a performance spersective, there's no difference. How you do your backups does matter however. If you use snapshots, a complete backup set is a snapshot of the log and database volumes. If you place the databases from two storage groups on the same volumes, and back up each storage group in a different backup set, then each backup set will contain the changed blocks from both storage gorups and presumably be twice it's normal size. Show quote "Mike O" <put_the_spam@the.can> wrote in message news:uB8AsKN$HHA.4476@TK2MSFTNGP06.phx.gbl... > Thanks for the info. I figured that having one disk array with more > spindles would give the best disk I/O, but I was concerned that multiple > databases on the same set of disks would cause performance issues and > offset better I/O of a single array. > > Mike O. > > "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message > news:%23GQJDgI$HHA.748@TK2MSFTNGP04.phx.gbl... >> Exactly. The reason you will want to use two databases is in the event >> one fails your time to restore is half that of a 100 GB database. You >> could break it down into four databases if you want. >> -- >> Ed Crowley >> MVP - Exchange >> "Protecting the world from PSTs and brick backups!" >> >> "Mike O" <put_the_spam@the.can> wrote in message >> news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... >>> >>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>> news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >>>> With the number of drives you have, I would try one storage group. Two >>>> drives RAID-1 for the OS, two drives RAID-1 for the transction logs, >>>> and 12 drives in a RAID-1+0 for the databases, but the feasibility of >>>> that depends on the inventory of drives you have. >>> >>> With one storage group, but still use 2 databases? I know the Exchange >>> can handle big databases, but the BPA flags it if the database is over >>> 100GB. Besides, I really don't ever want to do an emergency or mailbox >>> restore on a 200GB database.. >>> >>> >>>> -- >>>> Ed Crowley >>>> MVP - Exchange >>>> "Protecting the world from PSTs and brick backups!" >>>> >>>> "Mike O" <put_the_spam@the.can> wrote in message >>>> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>>>> I'm looking for some suggestions regarding hard drive configuration >>>>> for an Exchange 2003 system. >>>>> >>>>> We're in the process of building a new Exchange 2003 server as a >>>>> replacement for an existing server. It's an HP ML570 G2 with four >>>>> 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but it's newer >>>>> than the original one. Also, the original one has gone through several >>>>> Windows and Exchange service packs, hot fixes, etc., and I'd like to >>>>> get a clean 2003 R2 O/S install. I'm going to build the new server >>>>> then migrate the mailboxes then decommission the old one. >>>>> >>>>> Here's the situation: The existing server holds about 2,500 >>>>> mailboxes. This is a mailbox server only, OWA and the internet >>>>> connectivity is handled by other servers. It is currently set up as >>>>> two storage groups, with one database each. Each database (with STM >>>>> file) is about 90GB. The current arrangement has each database on its >>>>> own drive RAID5 array and each of logs on its own 2 drive RAID1. >>>>> The replacement server has 14 drive bays. Two will have the O/S >>>>> (RAID1), so that leaves 12 for Exchange. I have available a mix of >>>>> 36, 72, and 146GB 15K U320 SCSI drives. My preferred arrangement is >>>>> a single storage group/single database on a 10 drive RAID1 for the >>>>> data and a 2 drive RAID1 for the logs. However, I'd like to keep the >>>>> database size under 100G, so that means two databases. >>>>> >>>>> This brings up several options: >>>>> >>>>> Should I go with one storage group with two info stores, or two SGs >>>>> with one database each? I seem to remember when we set these up the >>>>> recommended configuration for multiple databases was multiple storage >>>>> groups, but is that still the recommended solution? >>>>> >>>>> With multiple databases, should I have each database on its own RAID1 >>>>> array? I could do something like two 4 drive RAID1 arrays (one for >>>>> each database), then two 2 drive RAID1 for the logs, one for each SG >>>>> (if I use two SG's). >>>>> >>>>> If I created one big drive array and put both databases on it, would >>>>> the performance of having a single larger array with more spindles >>>>> make up for putting two databases on the same array? >>>>> >>>>> With two storage groups, what about both databases on one 8 drive >>>>> RAID1 array, and two separate two drive RAID1 arrays, one for each >>>>> set of logs? >>>>> >>>>> I know there's all kinds of variations on this, but I'm trying to get >>>>> the best options on limited hardware. Hopefully this won't be a long >>>>> term need; we've submitted in the 2008 budget for new hardware & >>>>> migration to Exchange 2007 (we put it in this year but the hardware >>>>> budget was cut). >>>>> >>>>> Any suggestions would be appreciated. >>>>> >>>>> Mike O. >>>>> >>>> >>>> >>> >> >> > Thanks for the info. You mention that MS recommends each database in it's
own group so that you can isolate the logs. In one of the other replies, Ed Crowley mentioned about breaking the user mailboxes into more databases. But then he also recommended one storage group. If I went with multiple databases, based on the Microsoft direction wouldn't I need multiple storage groups (and multiple sets of log drives)? I understand because of the IO pattern I could have a "shared" RAID groups for the databases, but if create multiple storage groups I that would reduce the drives available for the data array. So what I'm looking at now is two storage groups, one database per group. An 8 drive RAID1 for the data and two RAID1 pairs, one for each log group's LOG files. - Mike O, Show quote "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message news:OAOkcnV$HHA.4612@TK2MSFTNGP03.phx.gbl... > The IO pattern of exchange databases is random. If you put two databases > on the same set of spindles, it's still random. The only difference is an > increse in the load the spindles must support. > > The IO pattern for logs is sequential. If you put two sets of logs on the > same set of physical spindles, the pattern changes to one that is less > sequential and there are performance reprecussions. > > Stores within a storage group share the same set of log files. Each > storage group has a seperate set of logs. The current MS guidance for > Storage Group design is to start with the largest number, even if you only > have one store per store group. This minimizes contention for a given set > of log files. Log writes are the most user visible aspect of Exchange IO. > http://support.microsoft.com/kb/890699/en-us > > Let's start by considering only best practices without regard to the > number of spindles you have available. To provide fault tolerance, the OS > volume should be mirrored. Each set of log files should be on physically > seperate spindles. RAID 1. 10, or 0+1 is preferred for databases. RAID 5 > does not provide an optimal (or arguably even acceptable) tradeoff between > fault tolerance and performance. If you go with two storage groups, then > you have 3 RAID 1 mirrors (one for the os and one for each set of logs). > For the databases you could use 1 large RAID 10, or two smaller RAID 10s. > From a performance spersective, there's no difference. How you do your > backups does matter however. If you use snapshots, a complete backup set > is a snapshot of the log and database volumes. If you place the databases > from two storage groups on the same volumes, and back up each storage > group in a different backup set, then each backup set will contain the > changed blocks from both storage gorups and presumably be twice it's > normal size. > > > > > "Mike O" <put_the_spam@the.can> wrote in message > news:uB8AsKN$HHA.4476@TK2MSFTNGP06.phx.gbl... >> Thanks for the info. I figured that having one disk array with more >> spindles would give the best disk I/O, but I was concerned that multiple >> databases on the same set of disks would cause performance issues and >> offset better I/O of a single array. >> >> Mike O. >> >> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >> news:%23GQJDgI$HHA.748@TK2MSFTNGP04.phx.gbl... >>> Exactly. The reason you will want to use two databases is in the event >>> one fails your time to restore is half that of a 100 GB database. You >>> could break it down into four databases if you want. >>> -- >>> Ed Crowley >>> MVP - Exchange >>> "Protecting the world from PSTs and brick backups!" >>> >>> "Mike O" <put_the_spam@the.can> wrote in message >>> news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... >>>> >>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>> news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >>>>> With the number of drives you have, I would try one storage group. >>>>> Two drives RAID-1 for the OS, two drives RAID-1 for the transction >>>>> logs, and 12 drives in a RAID-1+0 for the databases, but the >>>>> feasibility of that depends on the inventory of drives you have. >>>> >>>> With one storage group, but still use 2 databases? I know the Exchange >>>> can handle big databases, but the BPA flags it if the database is over >>>> 100GB. Besides, I really don't ever want to do an emergency or mailbox >>>> restore on a 200GB database.. >>>> >>>> >>>>> -- >>>>> Ed Crowley >>>>> MVP - Exchange >>>>> "Protecting the world from PSTs and brick backups!" >>>>> >>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>>>>> I'm looking for some suggestions regarding hard drive configuration >>>>>> for an Exchange 2003 system. >>>>>> >>>>>> We're in the process of building a new Exchange 2003 server as a >>>>>> replacement for an existing server. It's an HP ML570 G2 with four >>>>>> 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but it's >>>>>> newer than the original one. Also, the original one has gone through >>>>>> several Windows and Exchange service packs, hot fixes, etc., and I'd >>>>>> like to get a clean 2003 R2 O/S install. I'm going to build the new >>>>>> server then migrate the mailboxes then decommission the old one. >>>>>> >>>>>> Here's the situation: The existing server holds about 2,500 >>>>>> mailboxes. This is a mailbox server only, OWA and the internet >>>>>> connectivity is handled by other servers. It is currently set up as >>>>>> two storage groups, with one database each. Each database (with STM >>>>>> file) is about 90GB. The current arrangement has each database on its >>>>>> own drive RAID5 array and each of logs on its own 2 drive RAID1. >>>>>> The replacement server has 14 drive bays. Two will have the O/S >>>>>> (RAID1), so that leaves 12 for Exchange. I have available a mix of >>>>>> 36, 72, and 146GB 15K U320 SCSI drives. My preferred arrangement is >>>>>> a single storage group/single database on a 10 drive RAID1 for the >>>>>> data and a 2 drive RAID1 for the logs. However, I'd like to keep the >>>>>> database size under 100G, so that means two databases. >>>>>> >>>>>> This brings up several options: >>>>>> >>>>>> Should I go with one storage group with two info stores, or two SGs >>>>>> with one database each? I seem to remember when we set these up the >>>>>> recommended configuration for multiple databases was multiple storage >>>>>> groups, but is that still the recommended solution? >>>>>> >>>>>> With multiple databases, should I have each database on its own RAID1 >>>>>> array? I could do something like two 4 drive RAID1 arrays (one for >>>>>> each database), then two 2 drive RAID1 for the logs, one for each SG >>>>>> (if I use two SG's). >>>>>> >>>>>> If I created one big drive array and put both databases on it, would >>>>>> the performance of having a single larger array with more spindles >>>>>> make up for putting two databases on the same array? >>>>>> >>>>>> With two storage groups, what about both databases on one 8 drive >>>>>> RAID1 array, and two separate two drive RAID1 arrays, one for each >>>>>> set of logs? >>>>>> >>>>>> I know there's all kinds of variations on this, but I'm trying to get >>>>>> the best options on limited hardware. Hopefully this won't be a >>>>>> long term need; we've submitted in the 2008 budget for new hardware & >>>>>> migration to Exchange 2007 (we put it in this year but the hardware >>>>>> budget was cut). >>>>>> >>>>>> Any suggestions would be appreciated. >>>>>> >>>>>> Mike O. >>>>>> >>>>> >>>>> >>>> >>> >>> >> > > The reason I suggest a single storage group is that you're limited in the
number of drives. For 2,500 users, the drive arrangement you're considering might be on the edge where you start to see performance problems, depending on how active the users are, and therefore I think that splitting the databases onto two arrays and the logs onto two mirror sets would probably be a good thing. If you have Blackberry Enterprise Server and a lot of users with Blackberry devices, then I think that it's likely you'll have disk performance problems with the drive count limitation you have. -- Show quoteEd Crowley MVP - Exchange "Protecting the world from PSTs and brick backups!" "Mike O" <put_the_spam@the.can> wrote in message news:eNPp7VX$HHA.1212@TK2MSFTNGP05.phx.gbl... > Thanks for the info. You mention that MS recommends each database in > it's own group so that you can isolate the logs. In one of the other > replies, Ed Crowley mentioned about breaking the user mailboxes into more > databases. But then he also recommended one storage group. If I went with > multiple databases, based on the Microsoft direction wouldn't I need > multiple storage groups (and multiple sets of log drives)? I understand > because of the IO pattern I could have a "shared" RAID groups for the > databases, but if create multiple storage groups I that would reduce the > drives available for the data array. > > So what I'm looking at now is two storage groups, one database per group. > An 8 drive RAID1 for the data and two RAID1 pairs, one for each log > group's LOG files. > > > - Mike O, > > "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message > news:OAOkcnV$HHA.4612@TK2MSFTNGP03.phx.gbl... >> The IO pattern of exchange databases is random. If you put two databases >> on the same set of spindles, it's still random. The only difference is >> an increse in the load the spindles must support. >> >> The IO pattern for logs is sequential. If you put two sets of logs on >> the same set of physical spindles, the pattern changes to one that is >> less sequential and there are performance reprecussions. >> >> Stores within a storage group share the same set of log files. Each >> storage group has a seperate set of logs. The current MS guidance for >> Storage Group design is to start with the largest number, even if you >> only have one store per store group. This minimizes contention for a >> given set of log files. Log writes are the most user visible aspect of >> Exchange IO. http://support.microsoft.com/kb/890699/en-us >> >> Let's start by considering only best practices without regard to the >> number of spindles you have available. To provide fault tolerance, the >> OS volume should be mirrored. Each set of log files should be on >> physically seperate spindles. RAID 1. 10, or 0+1 is preferred for >> databases. RAID 5 does not provide an optimal (or arguably even >> acceptable) tradeoff between fault tolerance and performance. If you go >> with two storage groups, then you have 3 RAID 1 mirrors (one for the os >> and one for each set of logs). For the databases you could use 1 large >> RAID 10, or two smaller RAID 10s. From a performance spersective, there's >> no difference. How you do your backups does matter however. If you use >> snapshots, a complete backup set is a snapshot of the log and database >> volumes. If you place the databases from two storage groups on the same >> volumes, and back up each storage group in a different backup set, then >> each backup set will contain the changed blocks from both storage gorups >> and presumably be twice it's normal size. >> >> >> >> >> "Mike O" <put_the_spam@the.can> wrote in message >> news:uB8AsKN$HHA.4476@TK2MSFTNGP06.phx.gbl... >>> Thanks for the info. I figured that having one disk array with more >>> spindles would give the best disk I/O, but I was concerned that multiple >>> databases on the same set of disks would cause performance issues and >>> offset better I/O of a single array. >>> >>> Mike O. >>> >>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>> news:%23GQJDgI$HHA.748@TK2MSFTNGP04.phx.gbl... >>>> Exactly. The reason you will want to use two databases is in the event >>>> one fails your time to restore is half that of a 100 GB database. You >>>> could break it down into four databases if you want. >>>> -- >>>> Ed Crowley >>>> MVP - Exchange >>>> "Protecting the world from PSTs and brick backups!" >>>> >>>> "Mike O" <put_the_spam@the.can> wrote in message >>>> news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... >>>>> >>>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>>> news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >>>>>> With the number of drives you have, I would try one storage group. >>>>>> Two drives RAID-1 for the OS, two drives RAID-1 for the transction >>>>>> logs, and 12 drives in a RAID-1+0 for the databases, but the >>>>>> feasibility of that depends on the inventory of drives you have. >>>>> >>>>> With one storage group, but still use 2 databases? I know the >>>>> Exchange can handle big databases, but the BPA flags it if the >>>>> database is over 100GB. Besides, I really don't ever want to do an >>>>> emergency or mailbox restore on a 200GB database.. >>>>> >>>>> >>>>>> -- >>>>>> Ed Crowley >>>>>> MVP - Exchange >>>>>> "Protecting the world from PSTs and brick backups!" >>>>>> >>>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>>> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>>>>>> I'm looking for some suggestions regarding hard drive configuration >>>>>>> for an Exchange 2003 system. >>>>>>> >>>>>>> We're in the process of building a new Exchange 2003 server as a >>>>>>> replacement for an existing server. It's an HP ML570 G2 with four >>>>>>> 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but it's >>>>>>> newer than the original one. Also, the original one has gone through >>>>>>> several Windows and Exchange service packs, hot fixes, etc., and I'd >>>>>>> like to get a clean 2003 R2 O/S install. I'm going to build the new >>>>>>> server then migrate the mailboxes then decommission the old one. >>>>>>> >>>>>>> Here's the situation: The existing server holds about 2,500 >>>>>>> mailboxes. This is a mailbox server only, OWA and the internet >>>>>>> connectivity is handled by other servers. It is currently set up as >>>>>>> two storage groups, with one database each. Each database (with STM >>>>>>> file) is about 90GB. The current arrangement has each database on >>>>>>> its own drive RAID5 array and each of logs on its own 2 drive RAID1. >>>>>>> The replacement server has 14 drive bays. Two will have the O/S >>>>>>> (RAID1), so that leaves 12 for Exchange. I have available a mix of >>>>>>> 36, 72, and 146GB 15K U320 SCSI drives. My preferred arrangement >>>>>>> is a single storage group/single database on a 10 drive RAID1 for >>>>>>> the data and a 2 drive RAID1 for the logs. However, I'd like to >>>>>>> keep the database size under 100G, so that means two databases. >>>>>>> >>>>>>> This brings up several options: >>>>>>> >>>>>>> Should I go with one storage group with two info stores, or two SGs >>>>>>> with one database each? I seem to remember when we set these up the >>>>>>> recommended configuration for multiple databases was multiple >>>>>>> storage groups, but is that still the recommended solution? >>>>>>> >>>>>>> With multiple databases, should I have each database on its own >>>>>>> RAID1 array? I could do something like two 4 drive RAID1 arrays (one >>>>>>> for each database), then two 2 drive RAID1 for the logs, one for >>>>>>> each SG (if I use two SG's). >>>>>>> >>>>>>> If I created one big drive array and put both databases on it, would >>>>>>> the performance of having a single larger array with more spindles >>>>>>> make up for putting two databases on the same array? >>>>>>> >>>>>>> With two storage groups, what about both databases on one 8 drive >>>>>>> RAID1 array, and two separate two drive RAID1 arrays, one for each >>>>>>> set of logs? >>>>>>> >>>>>>> I know there's all kinds of variations on this, but I'm trying to >>>>>>> get the best options on limited hardware. Hopefully this won't be >>>>>>> a long term need; we've submitted in the 2008 budget for new >>>>>>> hardware & migration to Exchange 2007 (we put it in this year but >>>>>>> the hardware budget was cut). >>>>>>> >>>>>>> Any suggestions would be appreciated. >>>>>>> >>>>>>> Mike O. >>>>>>> >>>>>> >>>>>> >>>>> >>>> >>>> >>> >> >> > So it sounds like my best option is two storage groups, each with one
database. I understand the need for each storage group to have it's own log drives, but what about having both databases on one bigger array instead of two smaller ones. Since the nature of the I/O is random anyway, I'm thinking that having the traffic spread out over more spindles might spread the load more efficiently. We do have a Blackberry enterprise server, but there's only about 60 users between the two databases (we have another 100 or so, but on the other server). Mike O. Show quote "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message news:O1nmbIa$HHA.5328@TK2MSFTNGP05.phx.gbl... > The reason I suggest a single storage group is that you're limited in the > number of drives. For 2,500 users, the drive arrangement you're > considering might be on the edge where you start to see performance > problems, depending on how active the users are, and therefore I think > that splitting the databases onto two arrays and the logs onto two mirror > sets would probably be a good thing. If you have Blackberry Enterprise > Server and a lot of users with Blackberry devices, then I think that it's > likely you'll have disk performance problems with the drive count > limitation you have. > -- > Ed Crowley > MVP - Exchange > "Protecting the world from PSTs and brick backups!" > > "Mike O" <put_the_spam@the.can> wrote in message > news:eNPp7VX$HHA.1212@TK2MSFTNGP05.phx.gbl... >> Thanks for the info. You mention that MS recommends each database in >> it's own group so that you can isolate the logs. In one of the other >> replies, Ed Crowley mentioned about breaking the user mailboxes into more >> databases. But then he also recommended one storage group. If I went >> with multiple databases, based on the Microsoft direction wouldn't I >> need multiple storage groups (and multiple sets of log drives)? I >> understand because of the IO pattern I could have a "shared" RAID groups >> for the databases, but if create multiple storage groups I that would >> reduce the drives available for the data array. >> >> So what I'm looking at now is two storage groups, one database per group. >> An 8 drive RAID1 for the data and two RAID1 pairs, one for each log >> group's LOG files. >> >> >> - Mike O, >> >> "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message >> news:OAOkcnV$HHA.4612@TK2MSFTNGP03.phx.gbl... >>> The IO pattern of exchange databases is random. If you put two >>> databases on the same set of spindles, it's still random. The only >>> difference is an increse in the load the spindles must support. >>> >>> The IO pattern for logs is sequential. If you put two sets of logs on >>> the same set of physical spindles, the pattern changes to one that is >>> less sequential and there are performance reprecussions. >>> >>> Stores within a storage group share the same set of log files. Each >>> storage group has a seperate set of logs. The current MS guidance for >>> Storage Group design is to start with the largest number, even if you >>> only have one store per store group. This minimizes contention for a >>> given set of log files. Log writes are the most user visible aspect of >>> Exchange IO. http://support.microsoft.com/kb/890699/en-us >>> >>> Let's start by considering only best practices without regard to the >>> number of spindles you have available. To provide fault tolerance, the >>> OS volume should be mirrored. Each set of log files should be on >>> physically seperate spindles. RAID 1. 10, or 0+1 is preferred for >>> databases. RAID 5 does not provide an optimal (or arguably even >>> acceptable) tradeoff between fault tolerance and performance. If you go >>> with two storage groups, then you have 3 RAID 1 mirrors (one for the os >>> and one for each set of logs). For the databases you could use 1 large >>> RAID 10, or two smaller RAID 10s. From a performance spersective, >>> there's no difference. How you do your backups does matter however. If >>> you use snapshots, a complete backup set is a snapshot of the log and >>> database volumes. If you place the databases from two storage groups on >>> the same volumes, and back up each storage group in a different backup >>> set, then each backup set will contain the changed blocks from both >>> storage gorups and presumably be twice it's normal size. >>> >>> >>> >>> >>> "Mike O" <put_the_spam@the.can> wrote in message >>> news:uB8AsKN$HHA.4476@TK2MSFTNGP06.phx.gbl... >>>> Thanks for the info. I figured that having one disk array with more >>>> spindles would give the best disk I/O, but I was concerned that >>>> multiple databases on the same set of disks would cause performance >>>> issues and offset better I/O of a single array. >>>> >>>> Mike O. >>>> >>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>> news:%23GQJDgI$HHA.748@TK2MSFTNGP04.phx.gbl... >>>>> Exactly. The reason you will want to use two databases is in the >>>>> event one fails your time to restore is half that of a 100 GB >>>>> database. You could break it down into four databases if you want. >>>>> -- >>>>> Ed Crowley >>>>> MVP - Exchange >>>>> "Protecting the world from PSTs and brick backups!" >>>>> >>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>> news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... >>>>>> >>>>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>>>> news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >>>>>>> With the number of drives you have, I would try one storage group. >>>>>>> Two drives RAID-1 for the OS, two drives RAID-1 for the transction >>>>>>> logs, and 12 drives in a RAID-1+0 for the databases, but the >>>>>>> feasibility of that depends on the inventory of drives you have. >>>>>> >>>>>> With one storage group, but still use 2 databases? I know the >>>>>> Exchange can handle big databases, but the BPA flags it if the >>>>>> database is over 100GB. Besides, I really don't ever want to do an >>>>>> emergency or mailbox restore on a 200GB database.. >>>>>> >>>>>> >>>>>>> -- >>>>>>> Ed Crowley >>>>>>> MVP - Exchange >>>>>>> "Protecting the world from PSTs and brick backups!" >>>>>>> >>>>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>>>> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>>>>>>> I'm looking for some suggestions regarding hard drive configuration >>>>>>>> for an Exchange 2003 system. >>>>>>>> >>>>>>>> We're in the process of building a new Exchange 2003 server as a >>>>>>>> replacement for an existing server. It's an HP ML570 G2 with >>>>>>>> four 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but >>>>>>>> it's newer than the original one. Also, the original one has gone >>>>>>>> through several Windows and Exchange service packs, hot fixes, >>>>>>>> etc., and I'd like to get a clean 2003 R2 O/S install. I'm going >>>>>>>> to build the new server then migrate the mailboxes then >>>>>>>> decommission the old one. >>>>>>>> >>>>>>>> Here's the situation: The existing server holds about 2,500 >>>>>>>> mailboxes. This is a mailbox server only, OWA and the internet >>>>>>>> connectivity is handled by other servers. It is currently set up >>>>>>>> as two storage groups, with one database each. Each database (with >>>>>>>> STM file) is about 90GB. The current arrangement has each database >>>>>>>> on its own drive RAID5 array and each of logs on its own 2 drive >>>>>>>> RAID1. >>>>>>>> The replacement server has 14 drive bays. Two will have the O/S >>>>>>>> (RAID1), so that leaves 12 for Exchange. I have available a mix of >>>>>>>> 36, 72, and 146GB 15K U320 SCSI drives. My preferred arrangement >>>>>>>> is a single storage group/single database on a 10 drive RAID1 for >>>>>>>> the data and a 2 drive RAID1 for the logs. However, I'd like to >>>>>>>> keep the database size under 100G, so that means two databases. >>>>>>>> >>>>>>>> This brings up several options: >>>>>>>> >>>>>>>> Should I go with one storage group with two info stores, or two SGs >>>>>>>> with one database each? I seem to remember when we set these up >>>>>>>> the recommended configuration for multiple databases was multiple >>>>>>>> storage groups, but is that still the recommended solution? >>>>>>>> >>>>>>>> With multiple databases, should I have each database on its own >>>>>>>> RAID1 array? I could do something like two 4 drive RAID1 arrays >>>>>>>> (one for each database), then two 2 drive RAID1 for the logs, one >>>>>>>> for each SG (if I use two SG's). >>>>>>>> >>>>>>>> If I created one big drive array and put both databases on it, >>>>>>>> would the performance of having a single larger array with more >>>>>>>> spindles make up for putting two databases on the same array? >>>>>>>> >>>>>>>> With two storage groups, what about both databases on one 8 drive >>>>>>>> RAID1 array, and two separate two drive RAID1 arrays, one for each >>>>>>>> set of logs? >>>>>>>> >>>>>>>> I know there's all kinds of variations on this, but I'm trying to >>>>>>>> get the best options on limited hardware. Hopefully this won't be >>>>>>>> a long term need; we've submitted in the 2008 budget for new >>>>>>>> hardware & migration to Exchange 2007 (we put it in this year but >>>>>>>> the hardware budget was cut). >>>>>>>> >>>>>>>> Any suggestions would be appreciated. >>>>>>>> >>>>>>>> Mike O. >>>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>> >>>>> >>>> >>> >>> >> > > The only thing that idea would buy you is better performance on the log
volume, but if you're devoting a mirror set to the logs, I don't believe that the logs will be a bottleneck. I believe that if you see any performance problem it will be in the database. I think your proposed architecture is a good place to start, and if you have disk performance problems you could add drives later. I don't see any better way to carve up your limited drive arrangement than the way I've already suggested, and it may well work fine for you. -- Show quoteEd Crowley MVP - Exchange "Protecting the world from PSTs and brick backups!" "Mike O" <put_the_spam@the.can> wrote in message news:uh4JFhe$HHA.3548@TK2MSFTNGP06.phx.gbl... > So it sounds like my best option is two storage groups, each with one > database. I understand the need for each storage group to have it's own > log drives, but what about having both databases on one bigger array > instead of two smaller ones. Since the nature of the I/O is random > anyway, I'm thinking that having the traffic spread out over more spindles > might spread the load more efficiently. > > We do have a Blackberry enterprise server, but there's only about 60 users > between the two databases (we have another 100 or so, but on the other > server). > > Mike O. > > "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message > news:O1nmbIa$HHA.5328@TK2MSFTNGP05.phx.gbl... >> The reason I suggest a single storage group is that you're limited in the >> number of drives. For 2,500 users, the drive arrangement you're >> considering might be on the edge where you start to see performance >> problems, depending on how active the users are, and therefore I think >> that splitting the databases onto two arrays and the logs onto two mirror >> sets would probably be a good thing. If you have Blackberry Enterprise >> Server and a lot of users with Blackberry devices, then I think that it's >> likely you'll have disk performance problems with the drive count >> limitation you have. >> -- >> Ed Crowley >> MVP - Exchange >> "Protecting the world from PSTs and brick backups!" >> >> "Mike O" <put_the_spam@the.can> wrote in message >> news:eNPp7VX$HHA.1212@TK2MSFTNGP05.phx.gbl... >>> Thanks for the info. You mention that MS recommends each database in >>> it's own group so that you can isolate the logs. In one of the other >>> replies, Ed Crowley mentioned about breaking the user mailboxes into >>> more databases. But then he also recommended one storage group. If I >>> went with multiple databases, based on the Microsoft direction wouldn't >>> I need multiple storage groups (and multiple sets of log drives)? I >>> understand because of the IO pattern I could have a "shared" RAID groups >>> for the databases, but if create multiple storage groups I that would >>> reduce the drives available for the data array. >>> >>> So what I'm looking at now is two storage groups, one database per >>> group. An 8 drive RAID1 for the data and two RAID1 pairs, one for each >>> log group's LOG files. >>> >>> >>> - Mike O, >>> >>> "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message >>> news:OAOkcnV$HHA.4612@TK2MSFTNGP03.phx.gbl... >>>> The IO pattern of exchange databases is random. If you put two >>>> databases on the same set of spindles, it's still random. The only >>>> difference is an increse in the load the spindles must support. >>>> >>>> The IO pattern for logs is sequential. If you put two sets of logs on >>>> the same set of physical spindles, the pattern changes to one that is >>>> less sequential and there are performance reprecussions. >>>> >>>> Stores within a storage group share the same set of log files. Each >>>> storage group has a seperate set of logs. The current MS guidance for >>>> Storage Group design is to start with the largest number, even if you >>>> only have one store per store group. This minimizes contention for a >>>> given set of log files. Log writes are the most user visible aspect of >>>> Exchange IO. http://support.microsoft.com/kb/890699/en-us >>>> >>>> Let's start by considering only best practices without regard to the >>>> number of spindles you have available. To provide fault tolerance, the >>>> OS volume should be mirrored. Each set of log files should be on >>>> physically seperate spindles. RAID 1. 10, or 0+1 is preferred for >>>> databases. RAID 5 does not provide an optimal (or arguably even >>>> acceptable) tradeoff between fault tolerance and performance. If you >>>> go with two storage groups, then you have 3 RAID 1 mirrors (one for the >>>> os and one for each set of logs). For the databases you could use 1 >>>> large RAID 10, or two smaller RAID 10s. From a performance spersective, >>>> there's no difference. How you do your backups does matter however. >>>> If you use snapshots, a complete backup set is a snapshot of the log >>>> and database volumes. If you place the databases from two storage >>>> groups on the same volumes, and back up each storage group in a >>>> different backup set, then each backup set will contain the changed >>>> blocks from both storage gorups and presumably be twice it's normal >>>> size. >>>> >>>> >>>> >>>> >>>> "Mike O" <put_the_spam@the.can> wrote in message >>>> news:uB8AsKN$HHA.4476@TK2MSFTNGP06.phx.gbl... >>>>> Thanks for the info. I figured that having one disk array with more >>>>> spindles would give the best disk I/O, but I was concerned that >>>>> multiple databases on the same set of disks would cause performance >>>>> issues and offset better I/O of a single array. >>>>> >>>>> Mike O. >>>>> >>>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>>> news:%23GQJDgI$HHA.748@TK2MSFTNGP04.phx.gbl... >>>>>> Exactly. The reason you will want to use two databases is in the >>>>>> event one fails your time to restore is half that of a 100 GB >>>>>> database. You could break it down into four databases if you want. >>>>>> -- >>>>>> Ed Crowley >>>>>> MVP - Exchange >>>>>> "Protecting the world from PSTs and brick backups!" >>>>>> >>>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>>> news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... >>>>>>> >>>>>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>>>>> news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >>>>>>>> With the number of drives you have, I would try one storage group. >>>>>>>> Two drives RAID-1 for the OS, two drives RAID-1 for the transction >>>>>>>> logs, and 12 drives in a RAID-1+0 for the databases, but the >>>>>>>> feasibility of that depends on the inventory of drives you have. >>>>>>> >>>>>>> With one storage group, but still use 2 databases? I know the >>>>>>> Exchange can handle big databases, but the BPA flags it if the >>>>>>> database is over 100GB. Besides, I really don't ever want to do an >>>>>>> emergency or mailbox restore on a 200GB database.. >>>>>>> >>>>>>> >>>>>>>> -- >>>>>>>> Ed Crowley >>>>>>>> MVP - Exchange >>>>>>>> "Protecting the world from PSTs and brick backups!" >>>>>>>> >>>>>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>>>>> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>>>>>>>> I'm looking for some suggestions regarding hard drive >>>>>>>>> configuration for an Exchange 2003 system. >>>>>>>>> >>>>>>>>> We're in the process of building a new Exchange 2003 server as a >>>>>>>>> replacement for an existing server. It's an HP ML570 G2 with >>>>>>>>> four 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but >>>>>>>>> it's newer than the original one. Also, the original one has gone >>>>>>>>> through several Windows and Exchange service packs, hot fixes, >>>>>>>>> etc., and I'd like to get a clean 2003 R2 O/S install. I'm going >>>>>>>>> to build the new server then migrate the mailboxes then >>>>>>>>> decommission the old one. >>>>>>>>> >>>>>>>>> Here's the situation: The existing server holds about 2,500 >>>>>>>>> mailboxes. This is a mailbox server only, OWA and the internet >>>>>>>>> connectivity is handled by other servers. It is currently set up >>>>>>>>> as two storage groups, with one database each. Each database >>>>>>>>> (with STM file) is about 90GB. The current arrangement has each >>>>>>>>> database on its own drive RAID5 array and each of logs on its own >>>>>>>>> 2 drive RAID1. >>>>>>>>> The replacement server has 14 drive bays. Two will have the O/S >>>>>>>>> (RAID1), so that leaves 12 for Exchange. I have available a mix >>>>>>>>> of 36, 72, and 146GB 15K U320 SCSI drives. My preferred >>>>>>>>> arrangement is a single storage group/single database on a 10 >>>>>>>>> drive RAID1 for the data and a 2 drive RAID1 for the logs. >>>>>>>>> However, I'd like to keep the database size under 100G, so that >>>>>>>>> means two databases. >>>>>>>>> >>>>>>>>> This brings up several options: >>>>>>>>> >>>>>>>>> Should I go with one storage group with two info stores, or two >>>>>>>>> SGs with one database each? I seem to remember when we set these >>>>>>>>> up the recommended configuration for multiple databases was >>>>>>>>> multiple storage groups, but is that still the recommended >>>>>>>>> solution? >>>>>>>>> >>>>>>>>> With multiple databases, should I have each database on its own >>>>>>>>> RAID1 array? I could do something like two 4 drive RAID1 arrays >>>>>>>>> (one for each database), then two 2 drive RAID1 for the logs, one >>>>>>>>> for each SG (if I use two SG's). >>>>>>>>> >>>>>>>>> If I created one big drive array and put both databases on it, >>>>>>>>> would the performance of having a single larger array with more >>>>>>>>> spindles make up for putting two databases on the same array? >>>>>>>>> >>>>>>>>> With two storage groups, what about both databases on one 8 drive >>>>>>>>> RAID1 array, and two separate two drive RAID1 arrays, one for >>>>>>>>> each set of logs? >>>>>>>>> >>>>>>>>> I know there's all kinds of variations on this, but I'm trying to >>>>>>>>> get the best options on limited hardware. Hopefully this won't >>>>>>>>> be a long term need; we've submitted in the 2008 budget for new >>>>>>>>> hardware & migration to Exchange 2007 (we put it in this year but >>>>>>>>> the hardware budget was cut). >>>>>>>>> >>>>>>>>> Any suggestions would be appreciated. >>>>>>>>> >>>>>>>>> Mike O. >>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>> >>>>>> >>>>>> >>>>> >>>> >>>> >>> >> >> > In an ideal world with in unlimited number of spindles, I would follow the
MS recommendations to the letter. If you have something less, then you need to consider tradeoffs. Ed mentioned one tradeoff; sacrifice a little contention for the logs in order to reduce log spindle count. Another tradeoff I discussed was putting all the databases on the same drive (dependent on backup type). Of course you coulde always provision more spindles as well. Show quote "Mike O" <put_the_spam@the.can> wrote in message news:eNPp7VX$HHA.1212@TK2MSFTNGP05.phx.gbl... > Thanks for the info. You mention that MS recommends each database in > it's own group so that you can isolate the logs. In one of the other > replies, Ed Crowley mentioned about breaking the user mailboxes into more > databases. But then he also recommended one storage group. If I went with > multiple databases, based on the Microsoft direction wouldn't I need > multiple storage groups (and multiple sets of log drives)? I understand > because of the IO pattern I could have a "shared" RAID groups for the > databases, but if create multiple storage groups I that would reduce the > drives available for the data array. > > So what I'm looking at now is two storage groups, one database per group. > An 8 drive RAID1 for the data and two RAID1 pairs, one for each log > group's LOG files. > > > - Mike O, > > "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message > news:OAOkcnV$HHA.4612@TK2MSFTNGP03.phx.gbl... >> The IO pattern of exchange databases is random. If you put two databases >> on the same set of spindles, it's still random. The only difference is >> an increse in the load the spindles must support. >> >> The IO pattern for logs is sequential. If you put two sets of logs on >> the same set of physical spindles, the pattern changes to one that is >> less sequential and there are performance reprecussions. >> >> Stores within a storage group share the same set of log files. Each >> storage group has a seperate set of logs. The current MS guidance for >> Storage Group design is to start with the largest number, even if you >> only have one store per store group. This minimizes contention for a >> given set of log files. Log writes are the most user visible aspect of >> Exchange IO. http://support.microsoft.com/kb/890699/en-us >> >> Let's start by considering only best practices without regard to the >> number of spindles you have available. To provide fault tolerance, the >> OS volume should be mirrored. Each set of log files should be on >> physically seperate spindles. RAID 1. 10, or 0+1 is preferred for >> databases. RAID 5 does not provide an optimal (or arguably even >> acceptable) tradeoff between fault tolerance and performance. If you go >> with two storage groups, then you have 3 RAID 1 mirrors (one for the os >> and one for each set of logs). For the databases you could use 1 large >> RAID 10, or two smaller RAID 10s. From a performance spersective, there's >> no difference. How you do your backups does matter however. If you use >> snapshots, a complete backup set is a snapshot of the log and database >> volumes. If you place the databases from two storage groups on the same >> volumes, and back up each storage group in a different backup set, then >> each backup set will contain the changed blocks from both storage gorups >> and presumably be twice it's normal size. >> >> >> >> >> "Mike O" <put_the_spam@the.can> wrote in message >> news:uB8AsKN$HHA.4476@TK2MSFTNGP06.phx.gbl... >>> Thanks for the info. I figured that having one disk array with more >>> spindles would give the best disk I/O, but I was concerned that multiple >>> databases on the same set of disks would cause performance issues and >>> offset better I/O of a single array. >>> >>> Mike O. >>> >>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>> news:%23GQJDgI$HHA.748@TK2MSFTNGP04.phx.gbl... >>>> Exactly. The reason you will want to use two databases is in the event >>>> one fails your time to restore is half that of a 100 GB database. You >>>> could break it down into four databases if you want. >>>> -- >>>> Ed Crowley >>>> MVP - Exchange >>>> "Protecting the world from PSTs and brick backups!" >>>> >>>> "Mike O" <put_the_spam@the.can> wrote in message >>>> news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... >>>>> >>>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>>> news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >>>>>> With the number of drives you have, I would try one storage group. >>>>>> Two drives RAID-1 for the OS, two drives RAID-1 for the transction >>>>>> logs, and 12 drives in a RAID-1+0 for the databases, but the >>>>>> feasibility of that depends on the inventory of drives you have. >>>>> >>>>> With one storage group, but still use 2 databases? I know the >>>>> Exchange can handle big databases, but the BPA flags it if the >>>>> database is over 100GB. Besides, I really don't ever want to do an >>>>> emergency or mailbox restore on a 200GB database.. >>>>> >>>>> >>>>>> -- >>>>>> Ed Crowley >>>>>> MVP - Exchange >>>>>> "Protecting the world from PSTs and brick backups!" >>>>>> >>>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>>> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>>>>>> I'm looking for some suggestions regarding hard drive configuration >>>>>>> for an Exchange 2003 system. >>>>>>> >>>>>>> We're in the process of building a new Exchange 2003 server as a >>>>>>> replacement for an existing server. It's an HP ML570 G2 with four >>>>>>> 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but it's >>>>>>> newer than the original one. Also, the original one has gone through >>>>>>> several Windows and Exchange service packs, hot fixes, etc., and I'd >>>>>>> like to get a clean 2003 R2 O/S install. I'm going to build the new >>>>>>> server then migrate the mailboxes then decommission the old one. >>>>>>> >>>>>>> Here's the situation: The existing server holds about 2,500 >>>>>>> mailboxes. This is a mailbox server only, OWA and the internet >>>>>>> connectivity is handled by other servers. It is currently set up as >>>>>>> two storage groups, with one database each. Each database (with STM >>>>>>> file) is about 90GB. The current arrangement has each database on >>>>>>> its own drive RAID5 array and each of logs on its own 2 drive RAID1. >>>>>>> The replacement server has 14 drive bays. Two will have the O/S >>>>>>> (RAID1), so that leaves 12 for Exchange. I have available a mix of >>>>>>> 36, 72, and 146GB 15K U320 SCSI drives. My preferred arrangement >>>>>>> is a single storage group/single database on a 10 drive RAID1 for >>>>>>> the data and a 2 drive RAID1 for the logs. However, I'd like to >>>>>>> keep the database size under 100G, so that means two databases. >>>>>>> >>>>>>> This brings up several options: >>>>>>> >>>>>>> Should I go with one storage group with two info stores, or two SGs >>>>>>> with one database each? I seem to remember when we set these up the >>>>>>> recommended configuration for multiple databases was multiple >>>>>>> storage groups, but is that still the recommended solution? >>>>>>> >>>>>>> With multiple databases, should I have each database on its own >>>>>>> RAID1 array? I could do something like two 4 drive RAID1 arrays (one >>>>>>> for each database), then two 2 drive RAID1 for the logs, one for >>>>>>> each SG (if I use two SG's). >>>>>>> >>>>>>> If I created one big drive array and put both databases on it, would >>>>>>> the performance of having a single larger array with more spindles >>>>>>> make up for putting two databases on the same array? >>>>>>> >>>>>>> With two storage groups, what about both databases on one 8 drive >>>>>>> RAID1 array, and two separate two drive RAID1 arrays, one for each >>>>>>> set of logs? >>>>>>> >>>>>>> I know there's all kinds of variations on this, but I'm trying to >>>>>>> get the best options on limited hardware. Hopefully this won't be >>>>>>> a long term need; we've submitted in the 2008 budget for new >>>>>>> hardware & migration to Exchange 2007 (we put it in this year but >>>>>>> the hardware budget was cut). >>>>>>> >>>>>>> Any suggestions would be appreciated. >>>>>>> >>>>>>> Mike O. >>>>>>> >>>>>> >>>>>> >>>>> >>>> >>>> >>> >> >> > Thank you both for your comments. I think I'm going to go with two storage
groups, each with one database. One dual 36G drive RAID1 for each group log files and an 8 72G drive RAID1, which will give me 288G usable for the databases (both on the array) The existing server is using a three drive RAID5 for each database, so the new structure should be at least better than that.. Show quote "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message news:efUR6eg$HHA.1168@TK2MSFTNGP02.phx.gbl... > In an ideal world with in unlimited number of spindles, I would follow the > MS recommendations to the letter. If you have something less, then you > need to consider tradeoffs. Ed mentioned one tradeoff; sacrifice a > little contention for the logs in order to reduce log spindle count. > Another tradeoff I discussed was putting all the databases on the same > drive (dependent on backup type). Of course you coulde always provision > more spindles as well. > > > "Mike O" <put_the_spam@the.can> wrote in message > news:eNPp7VX$HHA.1212@TK2MSFTNGP05.phx.gbl... >> Thanks for the info. You mention that MS recommends each database in >> it's own group so that you can isolate the logs. In one of the other >> replies, Ed Crowley mentioned about breaking the user mailboxes into more >> databases. But then he also recommended one storage group. If I went >> with multiple databases, based on the Microsoft direction wouldn't I >> need multiple storage groups (and multiple sets of log drives)? I >> understand because of the IO pattern I could have a "shared" RAID groups >> for the databases, but if create multiple storage groups I that would >> reduce the drives available for the data array. >> >> So what I'm looking at now is two storage groups, one database per group. >> An 8 drive RAID1 for the data and two RAID1 pairs, one for each log >> group's LOG files. >> >> >> - Mike O, >> >> "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message >> news:OAOkcnV$HHA.4612@TK2MSFTNGP03.phx.gbl... >>> The IO pattern of exchange databases is random. If you put two >>> databases on the same set of spindles, it's still random. The only >>> difference is an increse in the load the spindles must support. >>> >>> The IO pattern for logs is sequential. If you put two sets of logs on >>> the same set of physical spindles, the pattern changes to one that is >>> less sequential and there are performance reprecussions. >>> >>> Stores within a storage group share the same set of log files. Each >>> storage group has a seperate set of logs. The current MS guidance for >>> Storage Group design is to start with the largest number, even if you >>> only have one store per store group. This minimizes contention for a >>> given set of log files. Log writes are the most user visible aspect of >>> Exchange IO. http://support.microsoft.com/kb/890699/en-us >>> >>> Let's start by considering only best practices without regard to the >>> number of spindles you have available. To provide fault tolerance, the >>> OS volume should be mirrored. Each set of log files should be on >>> physically seperate spindles. RAID 1. 10, or 0+1 is preferred for >>> databases. RAID 5 does not provide an optimal (or arguably even >>> acceptable) tradeoff between fault tolerance and performance. If you go >>> with two storage groups, then you have 3 RAID 1 mirrors (one for the os >>> and one for each set of logs). For the databases you could use 1 large >>> RAID 10, or two smaller RAID 10s. From a performance spersective, >>> there's no difference. How you do your backups does matter however. If >>> you use snapshots, a complete backup set is a snapshot of the log and >>> database volumes. If you place the databases from two storage groups on >>> the same volumes, and back up each storage group in a different backup >>> set, then each backup set will contain the changed blocks from both >>> storage gorups and presumably be twice it's normal size. >>> >>> >>> >>> >>> "Mike O" <put_the_spam@the.can> wrote in message >>> news:uB8AsKN$HHA.4476@TK2MSFTNGP06.phx.gbl... >>>> Thanks for the info. I figured that having one disk array with more >>>> spindles would give the best disk I/O, but I was concerned that >>>> multiple databases on the same set of disks would cause performance >>>> issues and offset better I/O of a single array. >>>> >>>> Mike O. >>>> >>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>> news:%23GQJDgI$HHA.748@TK2MSFTNGP04.phx.gbl... >>>>> Exactly. The reason you will want to use two databases is in the >>>>> event one fails your time to restore is half that of a 100 GB >>>>> database. You could break it down into four databases if you want. >>>>> -- >>>>> Ed Crowley >>>>> MVP - Exchange >>>>> "Protecting the world from PSTs and brick backups!" >>>>> >>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>> news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... >>>>>> >>>>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>>>> news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >>>>>>> With the number of drives you have, I would try one storage group. >>>>>>> Two drives RAID-1 for the OS, two drives RAID-1 for the transction >>>>>>> logs, and 12 drives in a RAID-1+0 for the databases, but the >>>>>>> feasibility of that depends on the inventory of drives you have. >>>>>> >>>>>> With one storage group, but still use 2 databases? I know the >>>>>> Exchange can handle big databases, but the BPA flags it if the >>>>>> database is over 100GB. Besides, I really don't ever want to do an >>>>>> emergency or mailbox restore on a 200GB database.. >>>>>> >>>>>> >>>>>>> -- >>>>>>> Ed Crowley >>>>>>> MVP - Exchange >>>>>>> "Protecting the world from PSTs and brick backups!" >>>>>>> >>>>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>>>> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>>>>>>> I'm looking for some suggestions regarding hard drive configuration >>>>>>>> for an Exchange 2003 system. >>>>>>>> >>>>>>>> We're in the process of building a new Exchange 2003 server as a >>>>>>>> replacement for an existing server. It's an HP ML570 G2 with >>>>>>>> four 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but >>>>>>>> it's newer than the original one. Also, the original one has gone >>>>>>>> through several Windows and Exchange service packs, hot fixes, >>>>>>>> etc., and I'd like to get a clean 2003 R2 O/S install. I'm going >>>>>>>> to build the new server then migrate the mailboxes then >>>>>>>> decommission the old one. >>>>>>>> >>>>>>>> Here's the situation: The existing server holds about 2,500 >>>>>>>> mailboxes. This is a mailbox server only, OWA and the internet >>>>>>>> connectivity is handled by other servers. It is currently set up >>>>>>>> as two storage groups, with one database each. Each database (with >>>>>>>> STM file) is about 90GB. The current arrangement has each database >>>>>>>> on its own drive RAID5 array and each of logs on its own 2 drive >>>>>>>> RAID1. >>>>>>>> The replacement server has 14 drive bays. Two will have the O/S >>>>>>>> (RAID1), so that leaves 12 for Exchange. I have available a mix of >>>>>>>> 36, 72, and 146GB 15K U320 SCSI drives. My preferred arrangement >>>>>>>> is a single storage group/single database on a 10 drive RAID1 for >>>>>>>> the data and a 2 drive RAID1 for the logs. However, I'd like to >>>>>>>> keep the database size under 100G, so that means two databases. >>>>>>>> >>>>>>>> This brings up several options: >>>>>>>> >>>>>>>> Should I go with one storage group with two info stores, or two SGs >>>>>>>> with one database each? I seem to remember when we set these up >>>>>>>> the recommended configuration for multiple databases was multiple >>>>>>>> storage groups, but is that still the recommended solution? >>>>>>>> >>>>>>>> With multiple databases, should I have each database on its own >>>>>>>> RAID1 array? I could do something like two 4 drive RAID1 arrays >>>>>>>> (one for each database), then two 2 drive RAID1 for the logs, one >>>>>>>> for each SG (if I use two SG's). >>>>>>>> >>>>>>>> If I created one big drive array and put both databases on it, >>>>>>>> would the performance of having a single larger array with more >>>>>>>> spindles make up for putting two databases on the same array? >>>>>>>> >>>>>>>> With two storage groups, what about both databases on one 8 drive >>>>>>>> RAID1 array, and two separate two drive RAID1 arrays, one for each >>>>>>>> set of logs? >>>>>>>> >>>>>>>> I know there's all kinds of variations on this, but I'm trying to >>>>>>>> get the best options on limited hardware. Hopefully this won't be >>>>>>>> a long term need; we've submitted in the 2008 budget for new >>>>>>>> hardware & migration to Exchange 2007 (we put it in this year but >>>>>>>> the hardware budget was cut). >>>>>>>> >>>>>>>> Any suggestions would be appreciated. >>>>>>>> >>>>>>>> Mike O. >>>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>> >>>>> >>>> >>> >>> >> > > 8-72GB RAID-1? Or do you mean RAID-1+0?
-- Show quoteEd Crowley MVP - Exchange "Protecting the world from PSTs and brick backups!" "Mike O" <put_the_spam@the.can> wrote in message news:%23HEgUWl$HHA.4612@TK2MSFTNGP03.phx.gbl... > Thank you both for your comments. I think I'm going to go with two > storage groups, each with one database. One dual 36G drive RAID1 for each > group log files and an 8 72G drive RAID1, which will give me 288G usable > for the databases (both on the array) > > The existing server is using a three drive RAID5 for each database, so the > new structure should be at least better than that.. > > > "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message > news:efUR6eg$HHA.1168@TK2MSFTNGP02.phx.gbl... >> In an ideal world with in unlimited number of spindles, I would follow >> the MS recommendations to the letter. If you have something less, then >> you need to consider tradeoffs. Ed mentioned one tradeoff; sacrifice a >> little contention for the logs in order to reduce log spindle count. >> Another tradeoff I discussed was putting all the databases on the same >> drive (dependent on backup type). Of course you coulde always provision >> more spindles as well. >> >> >> "Mike O" <put_the_spam@the.can> wrote in message >> news:eNPp7VX$HHA.1212@TK2MSFTNGP05.phx.gbl... >>> Thanks for the info. You mention that MS recommends each database in >>> it's own group so that you can isolate the logs. In one of the other >>> replies, Ed Crowley mentioned about breaking the user mailboxes into >>> more databases. But then he also recommended one storage group. If I >>> went with multiple databases, based on the Microsoft direction wouldn't >>> I need multiple storage groups (and multiple sets of log drives)? I >>> understand because of the IO pattern I could have a "shared" RAID groups >>> for the databases, but if create multiple storage groups I that would >>> reduce the drives available for the data array. >>> >>> So what I'm looking at now is two storage groups, one database per >>> group. An 8 drive RAID1 for the data and two RAID1 pairs, one for each >>> log group's LOG files. >>> >>> >>> - Mike O, >>> >>> "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message >>> news:OAOkcnV$HHA.4612@TK2MSFTNGP03.phx.gbl... >>>> The IO pattern of exchange databases is random. If you put two >>>> databases on the same set of spindles, it's still random. The only >>>> difference is an increse in the load the spindles must support. >>>> >>>> The IO pattern for logs is sequential. If you put two sets of logs on >>>> the same set of physical spindles, the pattern changes to one that is >>>> less sequential and there are performance reprecussions. >>>> >>>> Stores within a storage group share the same set of log files. Each >>>> storage group has a seperate set of logs. The current MS guidance for >>>> Storage Group design is to start with the largest number, even if you >>>> only have one store per store group. This minimizes contention for a >>>> given set of log files. Log writes are the most user visible aspect of >>>> Exchange IO. http://support.microsoft.com/kb/890699/en-us >>>> >>>> Let's start by considering only best practices without regard to the >>>> number of spindles you have available. To provide fault tolerance, the >>>> OS volume should be mirrored. Each set of log files should be on >>>> physically seperate spindles. RAID 1. 10, or 0+1 is preferred for >>>> databases. RAID 5 does not provide an optimal (or arguably even >>>> acceptable) tradeoff between fault tolerance and performance. If you >>>> go with two storage groups, then you have 3 RAID 1 mirrors (one for the >>>> os and one for each set of logs). For the databases you could use 1 >>>> large RAID 10, or two smaller RAID 10s. From a performance spersective, >>>> there's no difference. How you do your backups does matter however. >>>> If you use snapshots, a complete backup set is a snapshot of the log >>>> and database volumes. If you place the databases from two storage >>>> groups on the same volumes, and back up each storage group in a >>>> different backup set, then each backup set will contain the changed >>>> blocks from both storage gorups and presumably be twice it's normal >>>> size. >>>> >>>> >>>> >>>> >>>> "Mike O" <put_the_spam@the.can> wrote in message >>>> news:uB8AsKN$HHA.4476@TK2MSFTNGP06.phx.gbl... >>>>> Thanks for the info. I figured that having one disk array with more >>>>> spindles would give the best disk I/O, but I was concerned that >>>>> multiple databases on the same set of disks would cause performance >>>>> issues and offset better I/O of a single array. >>>>> >>>>> Mike O. >>>>> >>>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>>> news:%23GQJDgI$HHA.748@TK2MSFTNGP04.phx.gbl... >>>>>> Exactly. The reason you will want to use two databases is in the >>>>>> event one fails your time to restore is half that of a 100 GB >>>>>> database. You could break it down into four databases if you want. >>>>>> -- >>>>>> Ed Crowley >>>>>> MVP - Exchange >>>>>> "Protecting the world from PSTs and brick backups!" >>>>>> >>>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>>> news:ON4jk7%23%23HHA.4184@TK2MSFTNGP05.phx.gbl... >>>>>>> >>>>>>> "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message >>>>>>> news:uVIVHo%23%23HHA.3848@TK2MSFTNGP05.phx.gbl... >>>>>>>> With the number of drives you have, I would try one storage group. >>>>>>>> Two drives RAID-1 for the OS, two drives RAID-1 for the transction >>>>>>>> logs, and 12 drives in a RAID-1+0 for the databases, but the >>>>>>>> feasibility of that depends on the inventory of drives you have. >>>>>>> >>>>>>> With one storage group, but still use 2 databases? I know the >>>>>>> Exchange can handle big databases, but the BPA flags it if the >>>>>>> database is over 100GB. Besides, I really don't ever want to do an >>>>>>> emergency or mailbox restore on a 200GB database.. >>>>>>> >>>>>>> >>>>>>>> -- >>>>>>>> Ed Crowley >>>>>>>> MVP - Exchange >>>>>>>> "Protecting the world from PSTs and brick backups!" >>>>>>>> >>>>>>>> "Mike O" <put_the_spam@the.can> wrote in message >>>>>>>> news:ub7rYj%23%23HHA.5164@TK2MSFTNGP05.phx.gbl... >>>>>>>>> I'm looking for some suggestions regarding hard drive >>>>>>>>> configuration for an Exchange 2003 system. >>>>>>>>> >>>>>>>>> We're in the process of building a new Exchange 2003 server as a >>>>>>>>> replacement for an existing server. It's an HP ML570 G2 with >>>>>>>>> four 3.0 Ghz processors, 4GB of RAM. It's not new hardware, but >>>>>>>>> it's newer than the original one. Also, the original one has gone >>>>>>>>> through several Windows and Exchange service packs, hot fixes, >>>>>>>>> etc., and I'd like to get a clean 2003 R2 O/S install. I'm going >>>>>>>>> to build the new server then migrate the mailboxes then >>>>>>>>> decommission the old one. >>>>>>>>> >>>>>>>>> Here's the situation: The existing server holds about 2,500 >>>>>>>>> mailboxes. This is a mailbox server only, OWA and the internet >>>>>>>>> connectivity is handled by other servers. It is currently set up >>>>>>>>> as two storage groups, with one database each. Each database >>>>>>>>> (with STM file) is about 90GB. The current arrangement has each >>>>>>>>> database on its own drive RAID5 array and each of logs on its own >>>>>>>>> 2 drive RAID1. >>>>>>>>> The replacement server has 14 drive bays. Two will have the O/S >>>>>>>>> (RAID1), so that leaves 12 for Exchange. I have available a mix >>>>>>>>> of 36, 72, and 146GB 15K U320 SCSI drives. My preferred >>>>>>>>> arrangement is a single storage group/single database on a 10 >>>>>>>>> drive RAID1 for the data and a 2 drive RAID1 for the logs. >>>>>>>>> However, I'd like to keep the database size under 100G, so that >>>>>>>>> means two databases. >>>>>>>>> >>>>>>>>> This brings up several options: >>>>>>>>> >>>>>>>>> Should I go with one storage group with two info stores, or two >>>>>>>>> SGs with one database each? I seem to remember when we set these >>>>>>>>> up the recommended configuration for multiple databases was >>>>>>>>> multiple storage groups, but is that still the recommended >>>>>>>>> solution? >>>>>>>>> >>>>>>>>> With multiple databases, should I have each database on its own >>>>>>>>> RAID1 array? I could do something like two 4 drive RAID1 arrays >>>>>>>>> (one for each database), then two 2 drive RAID1 for the logs, one >>>>>>>>> for each SG (if I use two SG's). >>>>>>>>> >>>>>>>>> If I created one big drive array and put both databases on it, >>>>>>>>> would the performance of having a single larger array with more >>>>>>>>> spindles make up for putting two databases on the same array? >>>>>>>>> >>>>>>>>> With two storage groups, what about both databases on one 8 drive >>>>>>>>> RAID1 array, and two separate two drive RAID1 arrays, one for >>>>>>>>> each set of logs? >>>>>>>>> >>>>>>>>> I know there's all kinds of variations on this, but I'm trying to >>>>>>>>> get the best options on limited hardware. Hopefully this won't >>>>>>>>> be a long term need; we've submitted in the 2008 budget for new >>>>>>>>> hardware & migration to Exchange 2007 (we put it in this year but >>>>>>>>> the hardware budget was cut). >>>>>>>>> >>>>>>>>> Any suggestions would be appreciated. >>>>>>>>> >>>>>>>>> Mike O. >>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>> >>>>>> >>>>>> >>>>> >>>> >>>> >>> >> >> > Yes, RAID 1+0. Show quote "Ed Crowley [MVP]" <curspice@mvpsnospam.org> wrote in message news:ukRF8wm$HHA.536@TK2MSFTNGP06.phx.gbl... > 8-72GB RAID-1? Or do you mean RAID-1+0? > -- > Ed Crowley > MVP - Exchange > "Protecting the world from PSTs and brick backups!" > > "Mike O" <put_the_spam@the.can> wrote in message > news:%23HEgUWl$HHA.4612@TK2MSFTNGP03.phx.gbl... >> Thank you both for your comments. I think I'm going to go with two >> storage groups, each with one database. One dual 36G drive RAID1 for >> each group log files and an 8 72G drive RAID1, which will give me 288G >> usable for the databases (both on the array) >> >> The existing server is using a three drive RAID5 for each database, so >> the new structure should be at least better than that.. >> >> >> "John Fullbright" <fjohn@donotspamenetappdotcom> wrote in message >> news:efUR6eg$HHA.1168@TK2MSFTNGP02.phx.gbl... >>> In an ideal world with in unlimited number of spindles, I would follow >>> the MS recommendations to the letter. If you have something less, then >>> you need to consider tradeoffs. Ed mentioned one tradeoff; sacrifice a >>> little contention for the logs in order to reduce log spindle count. >>> Another tradeoff I discussed was putting all the databases on the same >>> drive (dependent on backup type). Of course you coulde always provision >>> more spindles as well. >>> >>> >>> "Mike O" <put_the_spam@the.can> wrote in message >>> news:eNPp7VX$HHA.1212@TK2MSFTNGP05.phx.gbl... >>>> Thanks for the info. You mention that MS recommends each database in >>>> it's own group so that you can isolate the logs. In one of the other >>>> replies, Ed Crowley mentioned about breaking the user mailboxes into >>>> more databases. But then he also recommended one storage group. If I >>>> went with multiple d | |||||||||||||||||||||||
