Link to home
Start Free TrialLog in
Avatar of Mathew Muthama
Mathew MuthamaFlag for Kenya

asked on


My primary goal is performance and redundancy for Exchange 2010.All roles will be on one server and a two DAG node setup .Kindly advice the best raid setup and disk partition for OS,Databases and Log files. Cost is not an issue.

Avatar of David
Flag of United States of America image

Link to home
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
RAID 0+1 will give the best combination of performance and redundancy.  RAID 1+0 will give the best combination of redundancy and performance.  The difference is slight but measurable.
RAID 10 is probably best for performance and redundancy.  Since you have few servers, I would stick with RAID 10.

You could do RAID 10 on the main server and RAID 0 on the DAGs since they are already redundant.
Performance is achieved with whatever level the storage calculator suggests. It is not as disk-bound as it was in previous versions of Exchange. In other words, if RAID-5 suffices, it is unlikely that you will see an improvement by going to RAID 0+1 (RAID10).

However, splitting users to multiple databases and therefore increasing the overall log checkpoint depth would insure more data is kept in RAM. This in turn would significantly improve performance.

Finally, if cost is not an option, you should consider going to more than two copies per database. For example, a 3-node DAG (perhaps with one server in a warm data centre) with RAID5 would give you far more redundancy than a 2-node DAG with RAID10.

In short, you should insert your user profile data into Exchange mailbox storage calculator, increase the number of servers, add RAM and implement what is suggested in a multi-role setup.

Probably best to confirm understanding before making recommendations.

"All roles will be on one server and a two DAG node setup "

Are you saying there will be a single server with the ewntire thing on, or a main server with all roles on and an additonal 2 Mailbox servers as part of the DAG (meaning 3 x mailbox roles in total)?

In gerenal I'd recommend a 2 disk mirror for the OS on all servers (hi speed disks where possible)

If cost isnt a problem a separate raid mirror for the logs is good, if not you can put them on the same disk as the OS. You wont see much performance decrease as the logs as a straight sequential write process so doesnt caiuse much issue. Redundancy is fine too as long as the logs are on seperate physical disks to the database.

Databases can be on a mirror as they do have good read performance, but if the databases are going to get large you can manage a raid 5 with hot spare, which would give good resilience and is fine performance wise for exchange DBs.

If you can afford it, id recommend 2 seperate RAID controllers for the seperate RAID logical drives as performance can be higher in the layout. either way, avoid cheapo RAID cards.
Honestly with 2010 you can run 25000 (easily) users on a JBOD 7200 rpm setup. There is no need for raid anymore... Everything really depends on how many users you have and what kind of servers you can procure.

Last environment I worked on was a 6 node dag where every (!!) server was sized to have 25 000 users simultaneously on them without performance issues. It worked too!
all nested raid arrays are better that older versions as they add a number of better methods of say protection like raid 10 for instance allows 2 drives to fail but keep the array working
as others have said, with Exchange 2010 disk IO is usually not the limiting factor. As this is in a DAG, and presumably you also have a load balancer in front of the cas-array, I would probably use RAID1 for each LUN apart from the C:\ LUN which I would probably use RAID10 , 5, 6, 50 or 60.

The reason I would have higher availability for boot and application over the databases and logs is that the databases and logs are already protected by the DAG, "cleaning up"  after a catastrophic failure of an Exchange server is something I would prefer to avoid...

if cost was not an issue, I would use individual LUNS for

C:\ used for boot and Exchange Installation
each database
each databases logs

rather than use LUNS mapped to drive letters, I would use LUNS mapped to folders, this allows for growth beyond 11 databases.

having mailboxes spread across a large number of small databases multiple databases instead of a small number of large databases can reduce recovery time in the event of a catastrophic LUN failure, as it takes less time to re-sync/recover a smaller database than a larger database...
Avatar of Mathew Muthama


I've requested that this question be deleted for the following reason:

Got external consultant.
Just because you got an external consultant doesn't mean the question wasn't answered properly.  You got wonderful advice from experts, and they answered the question.  

You pretty much did the equivalent of ordering a meal and then sending it back because you are no longer hungry.  Points should be awarded.