mf@ircam.ircam.fr (Michel Fingerhut) (09/26/90)
We noticed a significant slow-down of our 5820 (bi-processor 5800) in response time after upgrading from Ultrix 3.1 (which supports a single processor) to Ultrix 4.0 (whose "kernel has been enhanced to support multiple processors executing kernel code"). The local DEC support's reply to my call was that it was a known "problem" and their advice was to turn off the 2nd CPU until a fix is found. I did this, and lo and behold, "ls" runs so much faster... I wonder if by turning the remaining CPU off I'll get the computing power DEC claims for the DS5820.
rosenblg@cmcl2.NYU.EDU (Gary J. Rosenblum) (09/27/90)
I haven't had that problem on my 5820 with Ultrix 4.0. I timed an ls in a HUGE directory with both cpu's on, and then with one off, and the times were comparable. Did the DEC people give you any indication as to the problem (or a patch to 4.0?) Gary
ndoduc@framentec.fr (Nhuan Doduc) (09/27/90)
In <1990Sep26.151338.14480@ircam.ircam.fr> mf@ircam.ircam.fr (Michel Fingerhut) writes: >.... >I did this, and lo and behold, "ls" runs so much faster... I wonder if by >turning the remaining CPU off I'll get the computing power DEC claims for >the DS5820. You have 2 proc. you shut off 1 and get (my interpretation) twice (faster) My deduction is that if you shut down the remaining then the computing power would be infinitely faster. Hey look at this: 1 / half = twice 1 / zero = infinity CQFD --nh Nhuan DODUC, Framentec-Cognitech, Paris, France, ndoduc@framentec.fr or ndoduc@cognitech.fr, Association Francaise des Utilisateurs d'Unix, France, doduc@afuu.fr
mf@ircam.ircam.fr (Michel Fingerhut) (09/28/90)
It is really weird, then. The 5820 is *definitely* slower with 2 cpus. For instance, an "ls -l" of a local (i.e., non-NFS) directory containing 3 files and one sub-directory, on a machine with 2 users, takes 4-5 seconds (would you believe it). With one cpu off, it's instantaneous (what would happen with the second cpu off I wonder). DEC told me that it was a KNOWN problem, that they had experienced it too. So maybe it's a "feature" for Ultrix sent to non-US places? Moreover, they added that the CPU(s) and the disks (RA90) required some patches to the firmware which would not solve the problem, but which were required under Ultrix 4.0. Finally, one person at DEC told me that the 58n0 was not a success, being a bastardized machine in between VAXen and MIPS, expensive and slow. That person added that while benchmarks on the 5000 and the 5400 had shown even better results than predicted, it was definitely not the case on the 58n0. I'd be very interested in any experience (positive or negative) people in netland could report to me (and my direction). Thanks in advance...!