Poor VM performance

[goapps.us]

Hello everyone

We have been fighting some odd VM networking issues with our 2012R2 HV cluster. To troubleshoot, I actually isolated one host and created 2 VM's, I then created a private vswitch and connected my 2 test VM's.

Here are the results I'm getting on a old NON UCS HV host. The host is an old PowerEdge 300. It's a single processor server with 4 cores... seriously It's a POS.... https://www.evernote.com/l/ADnHLCEkuq9CrYZAZmfX3PDBcrnIMjjpzy0

What has me concerned are the results of my test VM's on my UCS host https://www.evernote.com/l/ADmHINDsAcJOhqvFbH2NVq5Jh7X7ucit-0Q. 

How is it possible for the VM's on our new B200M4 servers be under performing VM's on a 5y old m300?

Thanks

Note: To perform my tests, I'm using TamoSoft's free test utility

[Philip D'Ath]

I have had so many problems over the years with Hyper-V and network performance.  Nearly always the problems are resolved by turning off all the networking smarts on both the physical hosts and the virtual machines.  Basically follow this guide and turn everything off.

http://www.rackspace.com/knowledge_center/article/disabling-tcp-offloading-in-windows-server-2012

In that doesn't resolve it, try upgrading the NIC drivers on the physical host.

The base VMWare package for three servers is only about USD$500.  Make the change.  You wont regret it.  The cost of ongoing Hyper-V support is just too high.

http://store.vmware.com/store/vmware/en_US/pd/productID.282883900

[Philip D'Ath]

ps. You don't get these problems on older servers because they don't support any of the more advance hardware acceleration offload functions.  Hence why dumbing down new servers often resolves performance issues.

[goapps.us]

Hello and thank you for your reply. 

VMware VS Hyper-v is really not up to me here.. I have to identify the cause of this performance discrepancy and address it. That being said, my VM's are attached to a private switch, not associated with any physical adapters. So any advanced hardware offloading should not be at play here. Also, all the VM's were created with the same VHD image that was sysprep'ed. 

Thanks 

[Kirk J]

Greetings.

Can you tell me which processor model is in use and the blade firmware package installed?

Thanks,

Kirk...

[goapps.us]

Hello Kirk 

The servers are B200M4, processors are E5-2630V3 and the firmware is 2.2.4C. See complete details below. 

https://www.evernote.com/l/ADlzRqj-xvhMP4--b1CGMDobZjzP7bh3SBg

http://www.evernote.com/shard/s57/sh/ce10b87f-e6e4-4bd9-a44e-83db3ad2c3c3/0b0a73c240b6644d54cd3e564aeeec04

Thanks 

[Kirk J]

I was suspicious you might be hitting https://tools.cisco.com/bugsearch/bug/CSCuv04436/?reffering_site=dumpcr , but that should be resolved in 2.24c

Just to confirm, your blade's BIOS and CIMC have a version # that say 2.24c, and not just your overall UCSM firmware release?

Thanks,

Kirk..

[goapps.us]

Hello Kirk, I'm familiar with that bug... were were affected. See case 636356017 if you need any additional details . 

Based on my tests, there is a very significant difference between 2.24b and 2.25b. It ranges from 20% to 60% better performance.

Is Cisco certain that the bug is completely addressed with 2.24b? 

See more details on my results here

Thanks 

[goapps.us]

Hello Kirk 

I'm still trying to get the to the bottom of this performance issue. I'm not sure why this made a difference, but I deleted the adapters from my VM's and then recreated them. Based on my notes, all settings in the VM and on host are identical. 

However, results of the last test is completely different. Based on these results, the B200M4 is clearly outperforming the M300! 

Thanks