The reason for this is that the design of the databases in Exchange 2010 has 
been changed very much so from prior releases.

In prior releases you had a single table for messages, a single table for 
attachments, and so forth. This leads to A LOT of random I/O which is very 
expensive, especially on slower drives. In Exchange 2010, you have one table 
_per mailbox_ for folders, one per mailbox for message headers, and one per 
mailbox for bodies and so forth. What you get from this is huge gains in large 
sequential I/O because you have contiguous data.

The page size has also been increased to 32KB (4-fold increase) which gets you 
a substantial reduction in I/O operations. Say you have a 26KB message. That's 
4 reads in 2007, 1 read in 2010.

If you're running in a scenario where you have high availability (the new DAG 
[Distributed Availability Group] design), the checkpoint depth has been 
increased substantially which gets a huge reduction in write operations.

The build of Exchange 2010 that's available to the public now has a roughly 70% 
reduction in total IOPS per second. The numbers I'm looking at are for a 
scenario with 3000x250MB mailboxes with a very heavy user profile and 3MB of 
cache (RAM) allocated per user. That is a /huge/ performance gain.

The net result here is that you can buy fewer cheaper disks, put substantially 
more users on them (think potential factor of 3-4), and deliver larger mailbox 
quotas. That's a win.


Thanks,
Brian Desmond
br...@briandesmond.com

c - 312.731.3132


-----Original Message-----
From: Ben Scott [mailto:mailvor...@gmail.com]
Sent: Thursday, May 28, 2009 5:13 PM
To: NT System Admin Issues
Subject: Re: Amusing

On Thu, May 28, 2009 at 5:50 PM, Brian Desmond <br...@briandesmond.com> wrote:
> If 3,000 users all come after the same data in a short time window it's going 
> to be
> in the cache which means it's served out of memory (not disk). RAM is way
> faster than disk.

  Except that, from what you're saying, in Exchange 2010, there's no
SIS, so there will be no benefit from caching stuff in RAM, because
everybody has a unique copy of everything.

  Tell me again how this is an improvement?

-- Ben

~ Finally, powerful endpoint security that ISN'T a resource hog! ~
~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/>  ~


~ Finally, powerful endpoint security that ISN'T a resource hog! ~
~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/>  ~

Reply via email to