1 person found this helpful
SAP is a good partner of vmware and things work fairly well together. I haven't had the chance to start virtualizing our SAP instances yet, but what you describe is a good plan for high I/O high transactional systems. Putting one vm per LUN on a VMFS maximizes the I/O for that vm alone. Using RDM also isolates that vm, and allows you to use SAN-based snapshot/clone utilties which are fairly reliable, so that will work for you.
Other than that, I've found that the only way to get rid of these doubts is to run a test environment and actually do it. You know your environment, and for some systems, I never get the warm and fuzzy feeling that things will work correctly, until I do it. No matter how many helpful posts I may read.
I'm not sure about the one vmfs LUN per VM method, I was working on some benchmarks the other week and found that a single VM would only drive a governed amount of I/O based on the disk shares. Although this method would prevent scsi reservation issues. I was seeing a max of 30MB/s on a single VM when the ESX host was able to go upto 70MB/s outside the VM. I would have to bench it in that config to verify the same behavior. I would think it would still behave the same. RDM's are great for snapshots, I use them for large volumes greater than 100G, very nice point in time and LUN copy provisioning capabilities.
When I've put SAP on VMware, I've been mindful to adhere to SAP's own supported configuration stipulations and guidelines:
Lots of RAM per VM
Data disks as RDMs on RAID10 LUNs
Sap on Vmware is in a much better situation then people like Bovine and myself were in about a year ago! thankfully Oracle released a VM and SAP decided to get into bed with the big V
Anyway, when benchmarking with a run of SGEN on boxes it was very close on performance benefits between RDM and VMFS, what I was using was a great SAN (iSCSI) however it is all relative. I am sure newer 3.5 Large Page support will improve greatly here.
If you don't have interfering Basis teams then doing what Bovine suggests and tailoring SAP Best Practice is probably best.
This is good information, and I stand corrected. Thanks. I would love to see some more of your benchmarks, as I've seen in other posts that your methods are very meticulous, and I've been very impressed by the level of detail that you go into.
1 person found this helpful
Thats nice to hear,
Thanks for the compliment.
I will be doing more benching on how the shares impact a VM and how many high I/O VM's can be carried on a single vmfs store with FC vs iSCSI vs DASD.
I am finding that the swap files are not good on the SAN when the application is I/O intensive. When swap kicks in the disk performance in some cases is crippling. (Databases especially)
That sounds very interesting mike, ill see about different configuration about the number of luns and the possible perfomance inboth cases, the same lun with vmfs filesystem for some vm, or one per vm...
thanks a lot, great point of start