Comments on: Enterprise Computing: The Exchange Storage Bandwagon http://thestoragearchitect.com/2010/05/26/enterprise-computing-the-exchange-storage-bandwagon/ Storage, Virtualisation & Cloud Wed, 23 May 2012 02:49:52 +0000 hourly 1 http://wordpress.org/?v=3.2.1 By: hosted exchange service http://thestoragearchitect.com/2010/05/26/enterprise-computing-the-exchange-storage-bandwagon/#comment-1433 hosted exchange service Fri, 13 Aug 2010 21:13:45 +0000 http://www.thestoragearchitect.com/?p=1482#comment-1433 I think there's nothing wrong with overconfiguring. Sometimes it just helps to have that extra fast response times. I think there’s nothing wrong with overconfiguring. Sometimes it just helps to have that extra fast response times.

]]> By: John F. http://thestoragearchitect.com/2010/05/26/enterprise-computing-the-exchange-storage-bandwagon/#comment-1432 John F. Fri, 28 May 2010 01:43:42 +0000 http://www.thestoragearchitect.com/?p=1482#comment-1432 Oh, Exchange can be a beast at times. Back in Exchange 2003, short stroking wase in vogue. The IOPS/user were high, and things like Backberry or desktop search enginges (all the rage back then) drove them much higher. This is perhaps the genesis of overengineering the IOPS for Exchange to ensure adequate performance. Back then, a 50M mailbox was considered a good size. On to Exchange 2007, which really did drop the IOPS 70%. Some of it was by increasing the IO size from 4K to 8K, but the mojority was a reduction in Read IO due mainly to the larger database cache. It's not uncommon to see a 53:47 R/W ratio for cached Outlook clients. With the decrease in IO, mailbox sizes tended to grow. The 300-500M range was not uncommon. On to the latest and greatest; Exchange 2010. The cache increases again, with a few efficiency tricks. The IO size increases from 8K to 32K. Changes in log handling actually improve write performance. With DAG replication (log shipping) The IO ratio drifts closer to 60:40. The IOPS/user drops 70% again. If you're using mailboxes over about 700M, then you're in the SATA realm. Why is it I still see some vendors/consultants designing for Exchange 2003 when deploying Exchange 2010? When should you use SATA, and when is SAS more appropriate? How do you calculate the IO? What's the impact of Blackberry in Exchange 2007 and Exchange 2010? What's all this DAG stuff about? No, I don't expect you to have all the answers Chris, but I do believe you've inspired me to write a blog on the subject... Thanks J Oh, Exchange can be a beast at times. Back in Exchange 2003, short stroking wase in vogue. The IOPS/user were high, and things like Backberry or desktop search enginges (all the rage back then) drove them much higher. This is perhaps the genesis of overengineering the IOPS for Exchange to ensure adequate performance. Back then, a 50M mailbox was considered a good size.

On to Exchange 2007, which really did drop the IOPS 70%. Some of it was by increasing the IO size from 4K to 8K, but the mojority was a reduction in Read IO due mainly to the larger database cache. It’s not uncommon to see a 53:47 R/W ratio for cached Outlook clients. With the decrease in IO, mailbox sizes tended to grow. The 300-500M range was not uncommon.

On to the latest and greatest; Exchange 2010. The cache increases again, with a few efficiency tricks. The IO size increases from 8K to 32K. Changes in log handling actually improve write performance. With DAG replication (log shipping) The IO ratio drifts closer to 60:40. The IOPS/user drops 70% again. If you’re using mailboxes over about 700M, then you’re in the SATA realm.

Why is it I still see some vendors/consultants designing for Exchange 2003 when deploying Exchange 2010? When should you use SATA, and when is SAS more appropriate? How do you calculate the IO? What’s the impact of Blackberry in Exchange 2007 and Exchange 2010? What’s all this DAG stuff about? No, I don’t expect you to have all the answers Chris, but I do believe you’ve inspired me to write a blog on the subject…

Thanks

J

]]> By: Chris Evans http://thestoragearchitect.com/2010/05/26/enterprise-computing-the-exchange-storage-bandwagon/#comment-1431 Chris Evans Thu, 27 May 2010 11:31:38 +0000 http://www.thestoragearchitect.com/?p=1482#comment-1431 Tzvika See my post yesterday on Violin SSD - if you're using only 1/2 the drive to get throughput on spindle count, then perhaps you're using the wrong device.... Chris Tzvika

See my post yesterday on Violin SSD – if you’re using only 1/2 the drive to get throughput on spindle count, then perhaps you’re using the wrong device….

Chris

]]> By: Tzvika http://thestoragearchitect.com/2010/05/26/enterprise-computing-the-exchange-storage-bandwagon/#comment-1430 Tzvika Thu, 27 May 2010 09:00:05 +0000 http://www.thestoragearchitect.com/?p=1482#comment-1430 Sound advise. But I wouldn't argue 100% against short stroking, because there are many valid cases where it is extremely useful. Sometimes you just need the IOPS more than the space. Sound advise. But I wouldn’t argue 100% against short stroking, because there are many valid cases where it is extremely useful. Sometimes you just need the IOPS more than the space.

]]> By: Jim http://thestoragearchitect.com/2010/05/26/enterprise-computing-the-exchange-storage-bandwagon/#comment-1429 Jim Wed, 26 May 2010 21:56:39 +0000 http://www.thestoragearchitect.com/?p=1482#comment-1429 With Exchange 2010 it will be a lot easier to over configure the solution. People will still do it though. With Exchange 2010 it will be a lot easier to over configure the solution. People will still do it though.

]]>