iSCSI latency issue

[Willybobs27]

I have an implementation using Etherchannel VIFs for iSCSI with VMware that in the main works OK. However my a larger data set say 300GB is copied to a VM vCenter alerts for latency between 200 and 500ms for disk writes.

The NetApp shows no such latency through System Manager and sysstat -u shows CPU and Disk at a 60% maximum suggesting this is correct. ESXTOP does show the latency is real inside the VM.

The network people say the network is fine and not showing errors though iSCSI is shared on stacked Cisco 3750 switches.

Has anybody seen this before using such a configuration as per TR3749?

[AGREGERSEN]

I'm seeing a similiar issue with iSCSI, just using a physical SQL cluster. We also use VIF's and etherchannel (LACP). I opened a case on it (being new and all on NetApp), should hear something today - or so I was told. We are using Jumbo packets (9K) and 10gbe (the SQL's are unfortunately only 1gbe, so they use etherchannel as well). Using Extreme Networks switches.

Anders

[AGREGERSEN]

I was able to isolate my problem to a Windows setting. On Equallogic it didn't matter, but it made a huge difference using NetApp. I changed the autotuninglevel using netsh fra normal to disabled.

netsh interface tcp set global autotuninglevel=disabled

Anders

[DAVE_WITHERS]

Anders,

Thanks for posting your solution.  After peeling back layers of bugs and other issues on our 2x 3240's and now dealing iscsi latency issues with our SQL2k8 cluster, your post on resolution so far seems to have had massive impact(in a good way) on our SQL iscsi performance.  We saw an immediate increase in performance.

Home