paul@phoenix.Princeton.EDU (Paul Lansky) (09/05/88)
We have just upgraded to Ultrix 2.0 from 1.1, and I discover that our disk throughput seems to be considerably slower. Here is configuration MicrovaxII, EmulexQD32, two Fuji Eagles. Under 1.1 and 2.0 kernel is modified to allow 16k MAXBSIZE file systems. Under 1.1 we were able to get about 600k bytes per second. Under 2.0 rate is down to 250k bytes per second. I didnt remake file systems under 2.0. This is serious problem since we need to do high-speed data acquisition. Is it us, or is it 2.0?
paul@phoenix.Princeton.EDU (Paul Lansky) (09/05/88)
sorry, I left my address off previous posting on ultrix 2.0 disk throughput Paul Lansky Music Department Princeton University paul@phoenix.princeton.edu
pavlov@hscfvax.harvard.edu (G.Pavlov) (09/05/88)
In article <3591@phoenix.Princeton.EDU>, paul@phoenix.Princeton.EDU (Paul Lansky) writes: > We have just upgraded to Ultrix 2.0 from 1.1, and I discover that our > disk throughput seems to be considerably slower. Here is configuration > MicrovaxII, EmulexQD32, two Fuji Eagles. ...... > > This is serious problem since we need to do high-speed data acquisition. > Is it us, or is it 2.0? The problem is the QD32 controller. The only solution is to replace it with someone else's.You may want to try to obtain restitution from Emulex. Good luck. We finally did, but it required a massive effort: at least 40 hours on the phone, over a period of 5 months, involving at least 10 Emulex employees.. greg pavlov, fstrf, amherst, ny 716-834-0900
pavlov@hscfvax.harvard.edu (G.Pavlov) (09/06/88)
In article <615@hscfvax.harvard.edu>, pavlov@hscfvax.harvard.edu (G.Pavlov) writes: > The problem is the QD32 controller. The only solution is to replace it with > someone else's.You may want to try to obtain restitution from Emulex. Good > luck. We finally did, but it required a massive effort: at least 40 hours on > the phone, over a period of 5 months, involving at least 10 Emulex employees.. > Judging by mail to me, there is interest in more information on this. I can only tell you what I was told by Emulex; will keep it brief. The problem appears to be that DEC decreased the timeout interval on disk requests. The QD series emulates the MSCP protocol in microcode and does not reply to the cpu quickly enough, thus forcing frequent timeouts. The problem supposedly applies to all current QD series controllers. A new generation, to be available Reasonably Soon Now, is claimed by Emulex to be quick enough to overcome the problem. We are awaiting a set of Dilog 256 controllers as replacements. We have had one in-house for the past month for eveluation. It appears to be moderately faster than the Emulex QD32. But, as with the QD32, its per- formance (in our shop) with one disk is fine but so-so with two. > greg pavlov, fstrf, amherst, ny > 716-834-0900