Poor Autodeploy Performance on UCS B-Series

[peter.robertson]

"I have been rolling out Autodeploy to provision vSphere 5.0 U1 hosts in our environment.  We are running on Cisco UCS blades with full 10GbE connectivity within our datacenter fabric.  Our DHCP, DNS, TFTP and vCenter servers are all reachable over the 10GbE fabric, network and storage latency for these servers are well within tolerance and yet it takes 15-20 minutes to download the image and boot the server.  I realize that I am not pruning the ESX image, in fact I have taken the base image and added HA/FDM and Cisco vibs, but the image is still not huge.  The blades are running Intel Sandybridge with anywhere from 256-384 GB RAM.

I guess I just expected this to be faster.  It is considerably slower than booting from SAN.  Is my experience normal?"

I originally started this as a question on VMware's communities...  and the response I received from another user was that they were experiencing the same issue but only on their UCS blades, not other vendor's equipment.  So I pulled an old HP Proliant DL360 G5 off the shelf and ran a test.  The following times are from when the host starts the PXE process until ESX has finished booting:

HP G5 Rack  (1GbE, three switch hops)    < 3 minutes

UCS B200 M2/M3  (10GbE, 1 switch hop)    Anywhere from 12-20 minutes.

Has anyone else experienced this?  Any explanation as to why, or how to resolve?

Thanks!

[padramas]

Hello Peter,

Does the blades have Cisco VIC CNA ?

There is known issue in PXE implementation which is tracked by following defect.

http://tools.cisco.com/Support/BugToolKit/search/getBugDetails.do?method=fetchBugDetails&bugId=CSCtu17983

Padma

[peter.robertson]

Padma,

The M2s are using the Palo/M81KR CNA.  The M3s are using the 1240 LOM.  Both experience the problem.  The bug toolkit refers to the Palo experiencing problems.  Is it also to be expected with the 1240?

Thanks!

[padramas]

Peter,

I am not sure about it. I would suggest you to open a TAC Service Request with this defect id for further investigation.

Padma

[BIN67801BIN67800]

Hi Padma,

Are there any new insights on this issue? When will this issue get fixed?

Ingo

[aaronleech]

This has been plaguing us as well as we try to get our new UCS system going with ESX and autodeploy.

I see the bug status has changed to Fixed, but the release just says release pending.  Has anybody heard when this should be expected to be released?

[padramas]

Hello Aaron,

The defect is fixed in upcoming 2.1.2 release. It is tentatively scheduled for start of August.

Padma

[padramas]

Hello,

The issue is resolved in 2.1.2a which is available for download.

Thanks

Padma

[aaronleech]

I can confirm - we loaded the firmware last Wednesday, and boot times went from ~15 minutes, down to ~5 minutes.  Huge improvement.  I'm glad it finally got fixed!

[Boudewijn Plomp]

I notice this with WDS (Windows Deployment Services) as well. I use firmware version 2.1(3a). Although it does not take 10-15min as you mentioned, it takes longer than usual. About 5min as well. Doing this on another server with 10GB work within seconds!