Just a guess, but have you checked MTU sizes? Are you using a VSWITCH for the guests or dedicated OSA?
Scott Rohling On Wed, May 4, 2011 at 5:15 PM, Bhemidhi, Ashwin <ashw...@ti.com> wrote: > Hello all, > > > > Recently we started noticing on few of our Linux guest that Ethernet frames > were being delayed up to 25 seconds from the time they were sent to the time > the guest received them. The frames are Ethernet LLC keep alive polls > (layer 2 poll) that are sent by a Cisco SNA switch router every 30 secs. > Both the router and the Linux guest are in the same LAN. > > > > Looking at the ethereal traces captured on the guest. During normal > operation the keep alive Frames are being sent every 30 secs and the z/Linux > guest responds to the poll with in 60 micro seconds. But few times we > noticed that the frames were being delayed up to 25 seconds (total time from > the previous poll is 30+25) after the router sends the poll frame to the > time the Linux guest receives them. This is causing the keep alive timer ( > 9 secs = 1 sec X 8 retries) to expire and disconnect sessions. The Linux > guest eventually receives the frames including the retires all at the same > but by that time the sessions are dropped the router. It appears that the > frames are being buffered and are delayed by the guest receives them. > > > > We for sure know that the router is sending the poll every 30 seconds but > some were some how the frames were buffered (?) for 25 secs before being > delivered to the guest. I am trying to figure at which layer the delay was > being introduced. Are there any other traces that I can turn on z/VM to > diagnose the problem? Were do I start looking at? > > > > z/VM LPAR is a small one running 8 guest with 80MB memory and 16MB and 48MB > vdisk on a z10 > > > > IFL utilization : 2% X 2 IFLS, > > Central Storage : 95% 768 MB, > > XSTORE : 97% 256MB, > > PAGE : 12% X 2 3390-3 page DASD. > > > > Paging/Spooling activity: 0/s (most of the times) > > > > Thank you, > > Ashwin Bhemidhi >