I had a CPU pegging issue about 4.5 months ago on ESXi 5.0. I turned LRO off (only change I made), and I haven't had an issue with CPU pegging since.
My issue was due to the TCP throughput being irregular on VMXNET2 and VMXNET3 and it was causing HELLO/ACK issues with the database replication (it would get into a setup loop and never stop). Guest reboots didn't fix it, if a rebooted the UCS box it worked for a little bit and then went right back to the issue VMware states that the issue is known in Linux kernel 2.6.24 and later with VMXNET3 (http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1027511) There are plenty of reports on the Internet were the LRO/CUC issue has been found beyond esxi 4.1 and cucm 8.6. RHEL probably isn't going to get any performance gains from LRO anyhow so if it were me, I'd turn it off. It's broke now, right? Generally, you can't break broke so I would try it, if for no other reason than to say that isn't it. Again, I am not saying this IS your issue, just that it COULD BE a contributing factor, it would be worth a shot in my book if you haven't found the issue yet. From: wo...@justfamily.org Date: Mon, 17 Nov 2014 23:50:52 -0700 Subject: Re: [cisco-voip] VMware 5.5 To: ryanh...@outlook.com CC: m...@bilobit.com; james.buchan...@gmail.com; cisco-voip@puck.nether.net This discussion comes up about every 6 months on this list, LRO is no longer affected and was only a 4.1 issue on esxi. http://puck.nether.net/pipermail/cisco-voip/2012-October/029907.html http://docwiki.cisco.com/wiki/Unified_Communications_VMware_Requirements#Supported_Versions.2C_Patches_and_Updates_of_VMware_vSphere_ESXi The latter being definitive for me, they keep that doc updated pretty well and note, it states 4.1, not any other version of esxi. Martin has another issue. :) On Mon, Nov 17, 2014 at 4:46 PM, Ryan Huff <ryanh...@outlook.com> wrote: While that particular DocWiki does suggest that it is no longer needed to disable Large Receive Offset on ESXi 4.1 with CUCUM 8.6 and above; you should also note that one of the specific issues having LRO on can cause is consistent CPU pegging for the UCOS guests (wouldn't be the first time something didn't work the way Cisco's docs say it should). Please also review: https://supportforums.cisco.com/document/95886/disable-lro-ucs-uc-application-deployments Again, I can't say beyond a shadow of doubt that this is the issue anymore than anyone can say that it isn't. I would also look into the RAID drivers as well as all the physical connections on the UCS boxes. Thanks, rh From: wo...@justfamily.org Date: Mon, 17 Nov 2014 14:30:21 -0700 Subject: Re: [cisco-voip] VMware 5.5 To: ryanh...@outlook.com CC: m...@bilobit.com; james.buchan...@gmail.com; cisco-voip@puck.nether.net That page specifically says you don't have to disable LRO if you are above 4.1 esxi with8.6 CUCM. Martin, which version of 5.5 are you on? GA, update 1 or 2? I recently ran into an issue with a customer on 5.5 u1 and too new of raid drivers on the card and had to downgrade the drivers. Wasn't causing 100% cpu, but very slow drive access times. So checked your vmware version and compatibility on the hardware you are running. On Mon, Nov 17, 2014 at 11:59 AM, Ryan Huff <ryanh...@outlook.com> wrote: Martin, Have you disabled Large Receive Offset (LRO) on all the Elastic Sky hosts? http://docwiki.cisco.com/wiki/Disable_LRO Thanks, Ryan Huff CCIE Collaboration (Written), CCNP Voice, CCNA Voice, CCNA Route & Switch, CCNA Wireless UCCX Specialist From: m...@bilobit.com To: james.buchan...@gmail.com Date: Mon, 17 Nov 2014 18:06:51 +0000 CC: cisco-voip@puck.nether.net Subject: Re: [cisco-voip] VMware 5.5 So 5.5 is supported with CUCM 9.1.x So the question is: Why do I have CUCM at 100% CPU? Only hard reset helps at this point. Thanks, Martin From: James Buchanan [mailto:james.buchan...@gmail.com] Sent: Monday, November 17, 2014 1:10 PM To: Martin Schmuker Cc: Cisco VoIP Mailing List Subject: Re: [cisco-voip] VMware 5.5 Hello, Whenever you are looking for answers on UC virtualization, look here: http://docwiki.cisco.com/wiki/Virtualization_for_Cisco_Unified_Communications_Manager_%28CUCM%29 Thanks, James On Mon, Nov 17, 2014 at 5:36 AM, Martin Schmuker <m...@bilobit.com> wrote: Guys, since 3 weeks we are running our UC Environment (CUCM, Unity Cxn, IM&P) on ESXi 5.5. At Friday we setup new vCenter 5.5 and added the esx hosts. Since saturday, all machines are stuck in 100% CPU after a few hours. Sometimes all machines at the *same* time! They don’t reply Ping (ICMP Echo), and CPU is at 100%. CUCM and Cxn are on 9.1(2)SU2a (9.1.2.12901-3). Someone has any idea? Is 9.1.x not supported on vSphere 5.5? Thanks, Martin _______________________________________________ cisco-voip mailing list cisco-voip@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-voip _______________________________________________ cisco-voip mailing list cisco-voip@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-voip _______________________________________________ cisco-voip mailing list cisco-voip@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-voip
_______________________________________________ cisco-voip mailing list cisco-voip@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-voip
