[Iwg-arbitration-committee] Arbitration request for UNH failures.

[Rupert Dance]

Slava,

 

I am forwarding this to the arbitration committee as per the Logo Program
Policy.
<http://www.iol.unh.edu/services/testing/ofa/logoprogram/OFA-UNH-IOL_Logo_Pr
ogram-v1.14.pdf>  These requests will be reviewed and responded to by the
committee even though they will certainly be reviewing this with Nick.

 

Thanks

 

Rupert

 

From: Yaroslav Pekelis [mailto:slava at mellanox.com] 
Sent: Friday, February 03, 2012 3:53 PM
To: Nickolas Wood (ndv2 at iol.unh.edu); Rupert Dance (rsdance at soft-forge.com)
Cc: Amit Krig; Eyal Gutkind
Subject: FW: Arbitration request for UNH failures.

 

Nick,

 

I would like to fill official arbitration request on two issues:

 

First, link coming at SDR with Legacy Mellanox devices and ConnectX-3 based
FDR cards. 

Wrong rate was caused by simple configuration error that was injected into
INI of those cards between debug and logo events. The error was fixed in
later version and does not have any implication on the planned GA version of
the FW images. The proper images were provided to UNH immediately after
error was reported emphasizing the simplicity of the error and it
resolution.

 

Second, link coming at DDR speed with Qlogic 12200. 

For last two weeks I have done innumerous number of tests with 4 different
Qlogic 12200 switches and different cables. 

I did not test cables with more than 12DB attenuation loss - since this is
the attenuation the Qlogic switch supports - I did not see a single link
coming at other than QDR speed. 

In addition, deep analyze of the logs coming from the failing link shows
that QDR link negotiation was performed and then dismissed by the Spec
negotiation algorithms as having a lot of errors. This leads me to
conclusion that Qlogic switch used in the OFA cluster has been badly used
and may have ports in bad shape. This would explain why we have 3 other
links connected to SwitchX based devices MSX6036 and MSX6025 linked at
proper QDR speed. This also explains why I have proper link speed when using
switches I have in my lab.

 

Thus, I would like to ask arbitration committee to approve retest the
updated images for HCAs and redoing the testing of the link using cables
with better signal integrity and try more ports on both switches and if
possible another Qlogic switch.

 

Slava

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/iwg-arbitration-committee/attachments/20120203/4e64d4c2/attachment.html>