[comp.sys.sun] Which box is I/O hot ?

hart@tasman.cc.utas.edu.au (Bill Hart) (11/19/90)

Being fairly new to the net (3 months) and overloaded with work I haven't
set up a news system yet, so I'm borrowing a 'foreign' account to do this.
I work for a oceanographic research institute (CSIRO division of
oceanography), we are currently in the process of updating our very tired
and overworked Vax 11/750, to a Unix server. We expect to have between
20-30 people using the new machine, engaged in software development and
data processing. I expect that whatever machine we purchase it will be I/O
bound, so we wish to maximize our throughput. 

Our other requirments are being able to integrate the server with our
network of Sun/Solburne workstations, add on our own VME based disk/tape
controllers, and have an affordable upgrade path. The systems we have been
offered (which we can afford) are -

	Sony NEWS 3870 (25 MHz R3000) with 80MB + VME Slots

(I have a couple of questions about this machine ; is there an exabyte
driver for it?  is the memory SIMM memory of something else?  does it have
sync or async scsi ?)

	Mips M2000 with 32 MB 1 cpu
	SGI 210S 32 MB
	Solburne 5/601 32 MB 
	ICL DRS 6000 1 cpu 32 MB
	Sun 4/470 32MB 
	DG AV 5200 Single 25 MHz 88k 32MB
	Motorola Dual 20 MHz 88100 cpu 32 MB

It looks like we have been offered everything from huge boxes down to
headless workstations. I have done as much research as I can, but amongst
the mountains of glossies, spec sheets and benchmarks, nothing really
answers my question;

<<<<"Which machine, in a multiuser environment, with what disks and
controllers will have the best throughput ?" >>>>

Speaking to someone recently they told me that their recently purchased
IBM R6000 workstation had better throughput than their Sun 4/390 with IPI
drives. Mips claim that going from RISCos 3.0 to RISCos 4.0 they gained an
improvement in I/O performance of up to 600%. Comments like these leave me
with a small sense of despair, amost all of the above machines are
changing to SVR4 next year so if I do any benchmarking now will it mean
anything in 6 months.

The pain and anguish I am experiencing, is common to anybody making a
major computing purchase, unless you have budda like patience or a
perversion for being told blatant lies by sales reps. I don't have large
amounts of time to do extensive benchmarking and hand wringing so if
anybody can give me an idea of what machines amongst the above group have
have good throughput and any comparisons I would be most grateful. Any
comments on third party disks and controllers, software quality,
availability, hardware reliability etc would also be welcome. 

Please mail replies to me at hart@ocean.ml.csiro.au and I will send a
summary to the net. Thanks in advance 

Bill Hart.			Underpaid Scientist/Programmer/Network-Manager/
				Engineer/Interfacer-to-forked-tounged-salesreps
Email hart@ocean.ml.csiro.au	CSIRO Division of Oceanography, Australia
Ph (002) 206-446