We've seen something like this, not exactly the same. We have an open issue with IBM, but we're having trouble pinning it down. It originally started in zVM 4.2, and seems to be better in 4.3, but not completely resolved.
In 4.2, we had Linux instances on guest LAN, where SOME, but not all of them would consistently have terrible performance in and out. Changing IP addresses often moved or relieved the problem. IBM ultimately determined that packets were being dropped, but never found the cause. It did seem to be related to the number of hosts on a LAN though, and might have been aggravated by some misconfiguration of the guest LAN and TCPIP. We upgraded VM to 4.3, reconfigured the network somewhat, moved the Linux machines over intact, and the problem went (mostly) away. On 4.3, we haven't seen anything as serious, but we do occasionally see situations where a Linux guest will be running for a few days, and suddenly performance on the LAN connection will drop WAY off. Ping times, normally less than 1 ms., will increase to 500-1000 ms. or worse between guests on the SAME LAN. Usually, the problem is limited to one guest at a time, so, for example, with guests A, B, and C, A will be affected, so ping between A&B and A&C (both ways) will be slow, but ping between B&C will not. Rebooting the instance (without logging off) clears up the problem, leading me to believe it might be something in the qeth or qdio drivers. This DOESN'T seem to be related to the number of guests, more to traffic. It happens most often on our production LDAP server, and occasionally on production Samba servers. I've seen it on my test instance also, when I'm running ssh or telnet. The VM guys here have been trying to get IBM to provide some tools for monitoring the guest LAN. We've gotten some data using TCPDUMP on the guests, but the SuSE SLES8 drivers broke that. IBM provided a circumvention, and the fix for that will be in SP3, along with fixes to the qeth drivers for some other problems we found. > -----Original Message----- > From: Linux on 390 Port [mailto:[EMAIL PROTECTED] Behalf Of > Marcy Cortes > Sent: Tuesday, October 14, 2003 1:07 PM > To: [EMAIL PROTECTED] > Subject: Re: [LINUX-390] Ring Around The Guest Lan > > > I have 17 connected to one right now and haven't seen > anything like you're > seeing either. Do you have current maintenance on z/VM? What is the > MAXCONN setting on the guest lan (Q LAN DETAILS)? > > PS. What's BLAZE? > > Marcy Cortes > Wells Fargo Services Company > > -----Original Message----- > From: Post, Mark K [mailto:[EMAIL PROTECTED] > Sent: Tuesday, October 14, 2003 09:36 > To: [EMAIL PROTECTED] > Subject: Re: [LINUX-390] Ring Around The Guest Lan > > > Ann, > > I have 8+ SLES7 Linux/390 guests hooked to one Guest LAN on a > test box, and > have not seen anything like that. Sorry. > > > Mark Post > > -----Original Message----- > From: Ann Smith [mailto:[EMAIL PROTECTED] > Sent: Tuesday, October 14, 2003 12:33 PM > To: [EMAIL PROTECTED] > Subject: Ring Around The Guest Lan > > > We implemented a Guest Lan on our linux POC machine last week. > We are running 1 linux guest with WBIMB 5.0 (MQSI) , 1 guest > with Blaze > persistence mgr and Oracle data base, 1 guest running Blaze > and MQ queue > mgr., > and 3 other Blaze guests. Anyway, when the 4th linux guest attempts to > bring up Blaze the strangeness begins. We start losing connectivity > randomly to the linux guests. The linux guests do not go down but > something is very wrong. You can watch the connectivity failures move > from guest to guest. They are back then they are gone. I am still > waiting for the Blaze folks to tell me what messages the 4th guests > received. The Blze startup seems to have failed. So far we took a VM > restart dump and sent it off to IBM. But I figured I'd ask if > anyone has > seen a similar situation. We only have 1 more week with our loaner > hardware for our linux POC. We are running z/VM 4.3 and SuSE SLES8 (31 > bit). >
