home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.sun.hardware
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!wupost!sdd.hp.com!news.cs.indiana.edu!lynx!zia.aoc.nrao.edu!rmilner
- From: rmilner@zia.aoc.nrao.edu (Ruth Milner)
- Subject: Re: How does the new 2000+MIP Sun server compare against a Cray/XMP?
- Message-ID: <1992Nov19.195021.10810@zia.aoc.nrao.edu>
- Reply-To: rmilner@zia.aoc.nrao.edu (Ruth Milner)
- Organization: National Radio Astronomy Observatory, Socorro NM
- References: <1992Nov12.223309.7567@zia.aoc.nrao.edu> <lg859oINNa9j@exodus.Eng.Sun.COM>
- Date: Thu, 19 Nov 92 19:50:21 GMT
- Lines: 32
-
- In article <lg859oINNa9j@exodus.Eng.Sun.COM> walter@zarquon.Eng.Sun.COM writes:
- >In article 7567@zia.aoc.nrao.edu, rmilner@zia.aoc.nrao.edu (Ruth Milner) writes:
- >>The 2.19 GIPS is
- >>an aggregate number of the performance of all 20 processors using the
- >>SPECrate_int92 benchmark.
- >
- >You can ask your sales rep for the performance brief for the details.
-
- Yes, I have it; this was in fact where I got the numbers.
-
- >Actually, the 2.19 GIPS is based on Dhrystone 1.1, as is traditional.
- >SPECrate_int92 gives a measure of total system throughput on the SPEC
- >CINT92 benchmarks
-
- Hmmm ... well, I don't have the brief right in front of me (I'm not at work),
- but I don't recall that the Performance Brief mentions Dhrystones. I arrived
- at a number almost exactly the same as the 2.19GIPS for 20 processors simply
- by extrapolating from SPECrate_int92. I believe integer performance was only
- quoted there up to 8 processors. Since it's a measure of throughput, it's
- reasonably fair to just double it and add a bit. Obviously with 20 processors,
- the bus I/O performance starts to become a factor, but I'd be surprised if
- any of these CPU benchmarks had any significant I/O component.
-
- Does anyone have any figures for how well the buses handle heavy I/O on a
- 20-processor system? Our primary application can have a CPU to I/O ratio of
- about 10 FLOPS per byte (roughly), and there is some concern over how well
- the I/O throughput (both to memory and devices) will scale.
-
- Thanks.
- --
- Ruth Milner NRAO/VLA Socorro NM
- Computing Division Head rmilner@zia.aoc.nrao.edu
-