From: MERC::"uunet!WKUVX1.BITNET!MacroMan" 29-MAY-1992 14:46:16.66 To: MACRO32@WKUVX1.BITNET CC: Subj: Re: MTFO... >Date sent: 29-MAY-1992 10:01:23 > > > We've got a nonlocal configuration which is as follows: > [VAX 6430]-(cibca)--[star coupler]--(hsc70)-{RA90s} > > Periodically all DECserver drop our VAX service and disconnect > all users. This occurs at times of extremely high load on the > system. Additionally, it has been noted that the HSC is maxed > out several times during the day. > > Field Service: Oh yeah, you're maxing out your HSC so you're > hanging. Buy more {xxx} > > Ehud Gavron: What should happen is that IOs should be pending > and queued, not a hang for many seconds. > > Data: LTDRIVER does its work at IPL 8. PADRIVER at 20. An > outage of 2.55 seconds or more would cause this problem. > > I still insist it's a hardware problem; all the CI cards (VAX,HSC) > have been changed. Can anyone out there tell me if they've seen > this before, if it was hard or soft, and what the solution (if > any) was? > > Ehud > >-- >Ehud Gavron (EG76) >gavron@Spades.ACES.COM I while ago I had the same problem with DECservers dropping users. This was on a 6410/8530 CI cluster with an HSC70 and RA90s. The problem did not seem to correspond to any particular load on the system and I know the HSC wasn't maxed. What I did know is that I just upgraded LAT to 5.4 the day before. DEC admitted numerous problems with LAT 5.4 (specifically LTDRIVER) on 6000 series machines and provided me with a patch that didn't work either. For some unknown political reason Babson can't afford any down-time so I didn't have the chance to dig in and find out what was really going on. I assume you are running a higher version of LAT than this, and the circumstances are different, but the problem was the same and here it was a software problem. The fastest solution was to drop down to a pre-LATmaster version of LAT. (weak but it worked) John John Shea Systems Manager Babson College Shea@Babson.Bitnet (internet address soon to show here!)