Is anyone running this configuration?
Dell M1000e + M710HDs + Intel x520 cards + M8024k switches + EQ San
We have ESXi 5.0 on the blades, and have having a large amount of trouble getting the ISCSI network running properly through the 10gb intel cards/switches.
This is the problem:.
Vmware is configured correctly (verified by Vmware) with two vmk ports. Each port is bound to a nic port of the x520 card.
Switches are configured with a LAG between the two, and the SAN correctly setup and pluged one port into each switch.
Everything works fine with everything running. I see two paths to each LUN from each blade, evenly across both controllers.
If I pull a switch out, things go bad. All IO stops for 30-60secs before it normally returns, however sometimes the IO simply doesn't ever return(one in 10 switch "failures")
I should expect one path to drop, but IO should continue without even a interruption as far as I'm concerned as one path should be unaffected.
I have a seperate setup using physical servers, boardcom 10gb cards, m8024f switches and EQ SANs and that works fine with all kinds of failure modes with NO interruption.
Why are the blade switches so different/bad?
I've had multiple people from Dell ProSupport and PSG look at this, but no-one seems to have any experience with their own switches (the m8024ks). They have changed their mind from "looks like a bug" through to "it's meant to do that" several times.
Did you implement an iSCSI-Heartbeat?
On your iSCSI vSwitch you set up 2 VMkernels and isolate each to a single NIC. Then you bind them to the (software) iSCSI initiator.
However, you also need an iSCSI-heartbeat vmkernel that does use both NICs and should have the lowest vmkernel number in that vswitch (and probably best to also give it the lowest IP).
So you would end up with 3 VMkernels: (example)
vmk2: iSCSI-Heartbeat: 10.10.10.100 (vmnic2 and vmnic3)
vmk3: iSCSI-1: 10.10.10.101 (vmnic2 only)
vmk4: iSCSI-2: 10.10.10.102 (vmnic3 only)
When you have this in place, test/emulate a switch failure again to see if it still has the problem you're seeing or not.
Member since 2003
Thank you for that idea. I've never had that suggested to me.
I will give that a go.
The ISCSI-Hearbeat vmk, does that need to be bound to the iscsi initiator?
No, it's there just to keep the iSCSI link alive (from the SAN's point of view) on either NIC.
Member since 2003