Hi,
We've recently changed our VMFS layout and are seeing higher latencies on the new VMFS from an ESX standpoint..
We used to have out datastores layed out so each VM's VMDK's were on a seperate datastore. we decided to change this as we had reached the maximum number of paths to the SAN and also to simplify things such as backups. We now have all disks for each VM in a smaller number of larger datastores (1TB). We have apporx 20 VM's per datastore which is less than we had in the datastores containing OS vmdk's in the previous design, which was up to approx 25.
We are now seeing latencies averaging from 10ms up to 40ms with spikes upto 300ms. I wanted to know why this would be. I'm not seeing any storage related errors in vmkernel log such as reservation conflicts or aborts. I've checked the queues on the HBA's and they seem fine. Is it just the fact that we have consolodated more I/O onto larger LUNS? The thing thast confusing me is the Disk Group on the EVA (spindle count) and the load on the VM's hasn't changed. The one thing we haven't doen is load balance the storage paths but we hadn't done that in our previous design and we had no latency issues.
Thanks in advance
We've recently changed our VMFS layout and are seeing higher latencies on the new VMFS from an ESX standpoint..
We used to have out datastores layed out so each VM's VMDK's were on a seperate datastore. we decided to change this as we had reached the maximum number of paths to the SAN and also to simplify things such as backups. We now have all disks for each VM in a smaller number of larger datastores (1TB). We have apporx 20 VM's per datastore which is less than we had in the datastores containing OS vmdk's in the previous design, which was up to approx 25.
We are now seeing latencies averaging from 10ms up to 40ms with spikes upto 300ms. I wanted to know why this would be. I'm not seeing any storage related errors in vmkernel log such as reservation conflicts or aborts. I've checked the queues on the HBA's and they seem fine. Is it just the fact that we have consolodated more I/O onto larger LUNS? The thing thast confusing me is the Disk Group on the EVA (spindle count) and the load on the VM's hasn't changed. The one thing we haven't doen is load balance the storage paths but we hadn't done that in our previous design and we had no latency issues.
Thanks in advance
