Hi,
Our platform is running in ESXi4.1U2. We've got about 26 ESXi connected to our netapp storage (FAS6240).
Datastores are NFS qtrees.
We host about 900 VM.
It seems to be unstable.
In ESXi log, we found :
Apr 2 19:20:57 vmkernel: 6:09:57:16.302 cpu20:4295)VSCSI: 2526: handle 8200(vscsi0:0):Reset [Retries: 0/0]
Apr 2 19:21:00 vmkernel: 6:09:57:19.388 cpu25:661520)WARNING: VSCSI: 3265: handle 8200(vscsi0:0):WaitForCIF: Issuing reset; number of CIF:2
Apr 2 19:21:00 vmkernel: 6:09:57:19.388 cpu25:661520)WARNING: VSCSI: 2293: handle 8200(vscsi0:0):Ignoring double reset
Apr 2 19:21:06 vmkernel: 6:09:57:26.139 cpu20:4295)VSCSI: 2326: handle 8200(vscsi0:0):Completing reset (0 outstanding commands)
Apr 2 19:21:56 vmkernel: 6:09:58:16.302 cpu25:661520)VSCSI: 2252: handle 8200(vscsi0:0):Reset request on FSS handle 17044937 (6 outstanding commands)
Apr 2 19:21:57 vmkernel: 6:09:58:16.302 cpu20:4295)VSCSI: 2526: handle 8200(vscsi0:0):Reset [Retries: 0/0]
Apr 2 19:22:00 vmkernel: 6:09:58:19.316 cpu24:661520)WARNING: VSCSI: 3265: handle 8200(vscsi0:0):WaitForCIF: Issuing reset; number of CIF:7
Our Netapp parameters setted by VSC are :
Net.TcpipHeapSize = 30
Net.TcpipHeapMax = 120
NFS.MaxVolumes = 64
NFS.HeartbeatMaxFailures = 10
NFS.HeartbeatFrequency = 12
NFS.HeartbeatTimeout = 5
Do you think that we should increase tcp windows size like it's describe in : http://communities.vmware.com/thread/330413
In the last post, we can see this preconisation made for oracle :
the TCP receive window size from the default of 64K to 256K.
a. “options nfs.tcp.recvwindowsize 262144”
b. The source of this recommendation came from NetApp TR-3557 "HP-UX NFS Performance with Oracle".
c. By increasing the default 4 x we are giving the protocol stack more room for the application.
Because of the number of VM and ESX we have, maybe we should follow this preconisation ?