Wondering if anyone is virtualizing their Exchange?
We want to go ahead with this and were wondering what type of hardware people are using?
We were thinking of having 1 ESX Host dedicated for the Exchange VM. We have about 13000 users.
We are going with Exchange 2003 as of right now, not sure why but I dont make that call.
Disk layout would be
LOG1/MDB1 - VM_1.vmdk
LOG2/MDB2 - VM_2.vmdk
...etc for the remaining disks.
Anyone done something similar? Also we were thinking of just regular VMFS, not RDM. However in previous years I thought it was recommeneded that RDM be used for Exchange. Any thoughts?
Cheers,
Message was edited by: oreeh
Reason: Shortened the title
Oliver Reeh
If you are going to dedicate 1 host to exchange, why virtualize it?
Second, how many disks will you have behind those VMDKs? With 13000 users I'd hope that you have at least 13000 IOPs capabilty or at least 130 disks....
--Matt
VCP, vExpert, Unix Geek
We want to have all new servers virtualized, we're trying to make it a Standard. Plus I belive the dedicated part was originally brought up by the Exchange Admins, once we can prove the host is being under utilized we can then put other vm's on it.
So far we only have 5 disks!!
Log1/MDB1 - 70gb/410gb - .vmdk
Log2/MDB2 - 70/410 - _1.vmdk
Log3/MDB3 - 70/410 - _2.vmdk
Log4/MDB4 - 70/410 - _3.vmdk
Log5/MDB5 - 70/410 - _4.vmdk
This is why I asked about RDM's and Exchange, heard that was better due to high I/O.
VMFS and RDM have about the same performance according to all the papers, tests and benchmarks.
5 disks is NOT enough to run your number of users.you need 20X the disks.
--Matt
VCP, vExpert, Unix Geek
Our environment has 1250 users. Would strongly recommend using Exchange 2007 if you going to virtualize since that will be easier to upgrade to Exchange 2010 in the future. We are using Exchange 2007 with CCR clustering. I agree with the others in that you are going to need more disks for storage groups. If your users are like ours they create a *&!@ load of email. Our storage group layout looks somthing like this:
C: OS, Exchange 30GB VMDK
😧 Page file 20GB VMDK
E: Storage Group1 Logs 50GB RDM
F: Storage Group1 Data 100GB RDM
G: Storage Group2 Logs 50B RDM
H: Storage Group2 Data 100GB RDM
Create as many storage groups as you need. Consider mount point if you are going to run out of drive letters. You can use VMDK or RDM's. Size carefully for growth. Our servers have 20 RDM's attached for storage!
Mike
In test we have 39 disks in our Filer, so when we create a Lun whether it be 1GB or 500GB's it is spread across all 39 disks.
It is the same in prod, however I think we only have 25 disks in the 1 aggregate.
Do I still need to worry about how many VMDK's I need? I should be OK to do 1 vmdK for 410GB right?
By Creating more VMDK's do I just make the VMDK's smaller? Like as of right now we are only in need of 5 x 410GB MDB drives.
So should I create 10 x 205MDB drives?
Look at it from an IOPS and DR scenario. You will get better performance if you break this into smaller storage groups and lessen your exposure. The reason we broke it down into smaller units was because 100GB's I believe is I believe the limit on a storage group. Each RDM corresponds to a a LUN
number such as
RDM 1 (Hard disk 3) vmhba1:0:2:0
RDM 2 (Hard disk 4) vmhba1:0:3:0
RDM 3 (hard disk 5) vmhba1:0:4:0
But the IOPS is from a physical standpoint. One 100GB drive spread across 30 + disk, or one 400GB drive spread across 30 disks really matter?
Over 30 disks 100GB vs 400GB probably does not matter. Our SAN configuration is different. We have RAID 5 sets configured with between 5-7 drives.
What kind of RAID configuration do you have on SAN?