We had a similar issue with our VMware farm on our M1000e. We opened a case with VMware and found out that default generic network drivers were being used on the ESX hosts. VMware had us swap the default drivers for a newer set for broadcom and intel nics and we immediately saw a huge decrease in latency on the iSCSI and LAN networks to/from the VMs. Creating VMs from our templates sped up by 3X.
I would recommend you open a ticket and see if you have the similar default network drivers on your ESX hosts. Sent from my iPhone On Dec 24, 2011, at 10:07 PM, Patrick Cable <[email protected]> wrote: > Hi -tech. Merry $holiday. > > So, here's a fun gift that's been under my tree over the last few > weeks. Maybe someone has run into this before - I hope so, because > "slowness" issues can be endlessly frustrating. > > I have a Dell M1000e blade chassis. I have 4 of the M6220 16-port > blades in the A and C positions, and 2 of the M6348's in the B > position. The chassis has two 10G uplinks in a LAG (A1 and A2 have the > uplinks). Since you cant stack the M6220 with the M6348, I have a > small fiber jumper running from A1 and A2 to B1 and B2 respectively. > All the M6220s are stacked. Yes, the LAG groups are configured > properly on both sides. > > I am seeing high latency from my VMware chassis to anywhere else on > the LAN. Occasionally ping times in excess of 100ms+ will pop up on > our monitoring system, then after a 2nd or 3rd probe they'll > disappear. Running "du -sh /home/designs" (a NFS share connected to > the upstream switches) on a VM within the chassis takes 44 minutes > (real 43m50.722s, user 5.493s, sys 51.063s) to complete. On a compute > node directly on the upstream switches, it's only 14 minutes (real > 13m33.610s, user 4.061s, sys 47.292s). /home/designs have ~500gb in a > ton of tiny files, so 14 minutes doesnt totally surprise me. 44 > minutes, however... > > None of the links are maxed out. None of the interfaces show any > errors. Netflow information from within VMware don't show saturation. > esxtop doesnt show a high cpu or memory use, so... it should be some > sort of network latency issue? "vmstat" tells me (during 'du' for > example) that my cpu is near 100% idle, I have nothing waiting, etc. > This seems correct. > > I unfortunately have no insight into the upstream switches, but the > LAN team has looked at them and said "looks good to us." I'm not so > certain it's a VMware issue or else I'd submit a ticket with them. > > Any of your thoughts or comments would be appreciated. What have I missed? > > - Patrick > _______________________________________________ > Tech mailing list > [email protected] > https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech > This list provided by the League of Professional System Administrators > http://lopsa.org/ _______________________________________________ Tech mailing list [email protected] https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech This list provided by the League of Professional System Administrators http://lopsa.org/
