vswitch is blocking network interfaces of VM host

[Walter Kießl]

Hi,

when I start a vswitch on my IntegrityVM host (4.2, hpux11.31-201003) then glance reports more than 7.000.000 InPackets a second on the VM hosts physical network interface. But throughput is less than 1 KB a second. Those packets can be seen whether there is a VM running on the host or not. Stopping the vswitch drops InPackets to 0. After starting the vswitch InPacket rate rises again.

I have 4 IntegrityVM hosts. All 4 installed equally. All 4 show this behavior.

Is this a normal behavior? Has anybody seen this too?

Thanks for your help!

Best Regards
Walter

[Anonymous]

Yes, it's normal. Starting vswitch switches the NIC to promiscuous mode which means that it stops automatically dropping packets not targeted to the NIC itself because the upper layers (the hpvm network driver) has to see the packets as they could be for one of the guests attached to the vswitch/NIC.

[Walter Kießl]

Hi,

I'd like to warm up discussion about this topic.

This issue seems to produce some performance bottlenecks on the VM hosts. So LAN team did a network tracing outside the VM host. And this trace show that only a few thousand packets are passing from the swith to the VM host.

So I think the promiscous mode on the interfaces can not be the cause of this issue.

Do anybody have an idea where this packets come from?

Thanks for your help!

Walter

PS: I have BL870ci2 with an APA lan configuration using a VC ethernet module in a c7000 enclosure.

[Eric SAUBIGNAC]

Bonjour Walter,

"PS: I have BL870ci2 with an APA lan configuration using a VC ethernet module in a c7000 enclosure."

==> BL870ci2 means PCI-X 10 GbE NIC ?

If so check APA release notes, page 19, the line before "Required Disk Space" :

"PCI-X 10 GbE Fiber cards (failover groups only)"

Don't know if it can help you ...

Also, did you made a test with a virtual switch not attached to an APA ppa ?

Eric

[Walter Kießl]

Thanks for this hint.

Despite of being configured by an HP Consultant a knowledgebase article points out that configuring APA with VirtualConnectEthernetModules in other than loadbalancing mode is not supported.

So I disabled APA completely and reconfigured NIC and ServiceGuard without use of APA.

But the problem with the very high InPacket rate still exists. Same behaviour than described above.

Some more help will be greatly appreciated.

Thanks in advance!

[Fabio Ettore]

Hi Walter,
I would re-consider and re-read what Stan posted. If your doubt is the reason why you are seeing so many InPackets dropped, it's normal.

>>>>>>>
"So I disabled APA completely and reconfigured NIC and ServiceGuard without use of APA.

But the problem with the very high InPacket rate still exists. Same behaviour than described above."
<<<<<<<

This just shows it doesn't depend on APA configuration.

If you really note performance problems then the problem should be investigated somewhere else.

Best regards,
Fabio

[Walter Kießl]

Hi Fabio,

thanks for the answer.

I believed what Stan wrote - until our network team traced only a few thousand packets comming from the switch to the VM host. The VM host seems to multiply these packets to a few millions.

Best Regards
Walter

[cjb_1]

It was a while ago but we did have a problem with what glance was reporting in regard to network stats, it was reporting garbage (v high packet values,intermittently, not consistently like you suggest) and the fault went away following our patching run.

It may be worth validating what you are seeing (as your network people seem to suggest it's 'within' the vm host).

Have you tested to confirm there is a real bottleneck?

FYI this was the glance package that sorted our issue.

B3701AA C.05.00.050 HP GlancePlus/UX Pak for 11.31

[cjb_1]

I should have said this was only seen within the hpvm environment..