Exchange performance

[ecksmen]

People always forget about the client PCs. If you have a 20Gb mailbox, you'll also have a 20GB ost file, unless you're running online mode. This will be the cause of most end user experience frustrations.

Exchange 2010 was designed to run on jbod in a dag now IO isn't critical. They really reduced the disk IO requirement and with multiple mailbox database copies single disk resliency isn't required (though I don't share this opinion). They also recommend 10MB/ram per mailbox on top of 4GB of ram so you'll be looking at about at least 5GB of ram for 100 mailboxes. I'd opt for no less than 8GB for 100 users but rams cheap so put in as much as possible.

 

[HungryHippos]

People always forget about the client PCs. If you have a 20Gb mailbox, you'll also have a 20GB ost file, unless you're running online mode. This will be the cause of most end user experience frustrations.

Exchange 2010 was designed to run on jbod in a dag now IO isn't critical. They really reduced the disk IO requirement and with multiple mailbox database copies single disk resliency isn't required (though I don't share this opinion). They also recommend 10MB/ram per mailbox on top of 4GB of ram so you'll be looking at about at least 5GB of ram for 100 mailboxes. I'd opt for no less than 8GB for 100 users but rams cheap so put in as much as possible.

True, the OST file size is definitely a factor, check this article:

http://technet.microsoft.com/en-us/library/ee832791.aspx

Look at the Item Counts per Folder section, and it tells you how the bigger OST files can negatively affect performance.

I would strongly consider a Hosted Exchange solution if you don't want to manage a server, and the subsequent backups, licensing, purchasing you would need to do to bring an Exchange 2010 server up to spec.

 

[DRZ]

I'm not particularly sold on virtualising the mailbox roles in EX2010, although it is supported (but not vMotion etc). I'd much rather keep those physical and use the available clustering techniques to give you your uptime.

Servers like the Dell R720XD were practically built for such a purpose and keeping the storage local and in a DAG rather than running the DBs from shared storage keeps things simple. By all means virtualise the CAS and HT roles, watching out for any multicast issues with upstream switches - although much beyond your scale I'd invest in something proper to do the load balancing (F5, NetScaler etc) rather than rely on the MSLB stuff.

In terms of real-world application, these techniques have applied from deployments as small as yours to 10,000+ mailbox applications.

You need to get those 2003 DB sizes down, you're really, really going to struggle with keeping your migration on schedule with all of the item-level corruption you're inevitably going to encounter. Probably not a massive issue at 100 mailboxes but it is pretty painful at 10,000 mailboxes.