km@emory.UUCP (Ken Mandelberg) (06/23/85)
Does anyone have any performance information about
running Unix (presumably Ultrix) on a Vax 8600?
Right now we run about 30 users each on two 780s runing BSD. Would
60 users run better on one 8600. The users are students and faculty
mostly in an edit/compile cycle, with a smaller number doing
nroff/troff and running some statistical packages.
On the surface, the 8600 would seem like a win. DEC rates it at
about 4 780s. However, this is presumably a VMS comparision, and
it is not clear to me if it still holds true for Unix.
Can you think of any circumstances where the two 780s would
outperform the 8600?
Ken Mandelberg
Emory University
Dept of Math and CS
Atlanta, Ga 30322
{akgua,sb1,gatech,decvax}!emory!km USENET
km@emory CSNET
km.emory@csnet-relay ARPANET
--
Ken Mandelberg
Emory University
Dept of Math and CS
Atlanta, Ga 30322
{akgua,sb1,gatech,decvax}!emory!km USENET
km@emory CSNET
km.emory@csnet-relay ARPANETjsz@bridge2.UUCP (06/28/85)
> Can you think of any circumstances where the two 780s would > outperform the 8600? Yes. When one Vax goes down. -- -- UUCP: ...!decwrl!glacier!bridge2!jsz USPS: Jay Zusman / Bridge Communications / 1345 Shorebird / Mt View CA 94043 (415) 969 4400 x295
peterb@pbear.UUCP (06/29/85)
Since the 8600 still uses the same I/O structure as the 780's, heavy
swap will slow a 8600 down. Think of it this way, where will 2 780's
connected by ethernet be slower than an 8600? The answer is in inter machine
conversation, so the converse shoul ~ be true also.
Piling on more users to an 8600 will slow I/O somewhat since the
unibus is still there, but cpu intensive processes will excell on the 8600.
disk to disk transfer's such as copying, etc will be just about as fast. The
kernel will run faster allocating resources and resolving scheduling, but
many problems will be limited to the speed of I/O, and this controls overall
throughput for edit sessions.
If your users are running through the edit/compile phase most of the
time, a possible step prior to aquiring ab 8600 is to try for the best of
all worlds, namely pile your user's terminals on one 780 and do the compiles
on the other. This way the I/O 780 is doing mostly slow I/O terminal work
without cpu intensive operations slowing down the I/O (such as compiles).
On the other 780, the cpu intensive work will not have the massive I/O
overhead of supporting terminals and their inherent speed. It will sit in
cpu intensive state most (if not all) of the time.
This situation presumes that you ethernet to two together and remove
the distinction from the user by placing shell scripts in fromt of the
compilers. It may not work perfectly, but average throughput should increase
as you can tune each machine for a different environment, therby maximizing
the throughput.
Hope it helps. If all else fails, find an 8600 and run your own
backyard benchmarks.
Peter Barada
{ihnp4!inmet|{harvard|cca}!ima}!pbear!peterbkeith@motel6.UUCP (Keith Packard) (06/29/85)
In article <1513@emory.UUCP> km@emory.UUCP (Ken Mandelberg) writes: >Right now we run about 30 users each on two 780s runing BSD. Would >60 users run better on one 8600. The users are students and faculty >mostly in an edit/compile cycle, with a smaller number doing >nroff/troff and running some statistical packages. > >On the surface, the 8600 would seem like a win. DEC rates it at >about 4 780s. However, this is presumably a VMS comparision, and >it is not clear to me if it still holds true for Unix. > > >Ken Mandelberg >Emory University >Dept of Math and CS >Atlanta, Ga 30322 Sites like this might want to look into systems like the sequent balance 8000 or other multi-cpu systems. We have had a sequent box for about 3 months and I, for one, would never consider buying a vax again in a multi-user environment. It's got 6 32016's and a mess of iop's and runs 4.2 unix. For single job execution it performs about like an 11/750. For 6 job execution it performs about like 6 11/750's. Even the i/o bandwidth doesn't seem to slow it down a bit, the notoriously cpu bound 4.2 file system has a party with 6 cpu's serving it! And, the best part, it costs less than a *single* 11/780! Also, it is housed in a rather small box (1m deep, 2m wide and <1m high). I use it for software development - edit, compile, link... and have been working with ~1M of code. I was on an 11/780 with about 30 other software designers, load averages of 20-40 not uncommon. The sequent box has been wonderful. They are supposedly coming out with 32032 boards that are plug compatible with the 32016 boards and run a bit faster. The days of single CPU's multitasking for a multitude of users are numbered, there are only two directions I see of change, either sun's on every desk or systems like the sequent box. I would rather have the sun, but the sequent box is more in keeping with traditional ideas of centrallized computing power, as well as being well suited for the university environment where a $17000 machine availible for general use would tend to be destroyed in less than a year. I, of course, have no connection with sequent except as a satisfied user. keith packard ...!tektronix!tekmdp!keithp (this is the sequent box) ...!tektronix!reed!keith (home)
oz@yetti.UUCP (Ozan Yigit) (07/04/85)
In article <7500004@pbear.UUCP> peterb@pbear.UUCP writes: > > > Since the 8600 still uses the same I/O structure as the 780's, heavy >swap will slow a 8600 down. > The only reason that an 8600 will slow down is because the 8600 is the only VAX that can push say an RA81 to its performance limit. > Piling on more users to an 8600 will slow I/O somewhat since the >unibus is still there, but cpu intensive processes will excell on the 8600. > Even an 8600 does not use up the entire unibus bandwidth. We found that our 8600 was on its knees (running VMS) because of a job with extremely high paging rate. It actually was due to an RA81 (there was only one disk at that time) becoming a bottleneck, NOT the unibus. Oz -- Oz [all wizardesque side effects are totaly unintentional, unless stated otherwise..] Usenet: [decvax | allegra | linus | ihnp4] !utzoo!yetti!oz Bitnet: oz@ [yuleo | yuyetti] ------------- Support GNU. Consider the 3-muskateers' motto: ONE FOR ALL - ALL FOR ONE