km@mathcs.emory.edu (Ken Mandelberg) (02/23/90)
I notice that the latest price list from Sun lists the IPI controller
as a separate item. Previously you had to buy a whole IPI subsystem.
Has anyone tried to mix and match 3rd party IPI disks with this
controller, either for economy or better performance? The rumor I hear
is that Sun uses the SABER drive which is rated 3MB/Sec, and doesn't
buy much over the best SMD and SCSI subsystems that are out.
--
Ken Mandelberg | km@mathcs.emory.edu PREFERRED
Emory University | {decvax,gatech}!emory!km UUCP
Dept of Math and CS | km@emory.bitnet NON-DOMAIN BITNET
Atlanta, GA 30322 | Phone: (404) 727-7963lamaster@ames.arc.nasa.gov (Hugh LaMaster) (02/28/90)
In article <5048@emory.mathcs.emory.edu> km@mathcs.emory.edu (Ken Mandelberg) writes: >is that Sun uses the SABER drive which is rated 3MB/Sec, and doesn't >buy much over the best SMD and SCSI subsystems that are out. Yes, the rumor is at least partially true, since at least *some* of the Sun disks are Sabre IPI disks. But, most companies, Sun included, reserve the right to change sources to handle supply/cost/reliability issues. Anyway, there is a good reason to use IPI, whatever the source: The throughput is much faster using IPI on Sun systems, anyway. Highly recommended for people with disk performance problems. Hugh LaMaster, m/s 233-9, UUCP ames!lamaster NASA Ames Research Center ARPA lamaster@ames.arc.nasa.gov Moffett Field, CA 94035 Phone: (415)604-6117
andrew@alice.UUCP (Andrew Hume) (02/28/90)
listen, if you have an ipi interface, do yourself a favour and get some real disks. For example, the Sabre 2HP is only a little more expensive than the regular sabre, a tad smaller but twice as fast.
lamaster@ames.arc.nasa.gov (Hugh LaMaster) (03/01/90)
In article <10538@alice.UUCP> andrew@alice.UUCP (Andrew Hume) writes: >listen, if you have an ipi interface, do yourself a favour >and get some real disks. For example, the Sabre 2HP is only >a little more expensive than the regular sabre, a tad smaller >but twice as fast. I assume that you mean 6 MB/sec. I guess Sun doesn't sell these. Are there any problems with drivers/formatting/etc? Can you just plug them in and run? I thought the filesystem was the bottleneck. The best you can get out of a 3 MB/sec disk is ~2MB/sec now. What kind of speeds can you get out of a 6 MB/sec disk? I believe that the maximum throughput to memory on a Sun-4/490 is 13.4MB/sec based on the paper by Hsieh et al. in last falls IEEE Computer Design. Do you have any insight into the aggregate performance improvement with faster disks? Hugh LaMaster, m/s 233-9, UUCP ames!lamaster NASA Ames Research Center ARPA lamaster@ames.arc.nasa.gov Moffett Field, CA 94035 Phone: (415)604-6117
andrew@alice.UUCP (Andrew Hume) (03/02/90)
i don't actually run Suns; but we were told some preliminary measurements from sgi. they get about 2MB/s thru the filesystem (actaully better but who's counting) on 3MB/s drives and this performance roughly scaled up with the 6MB/s drive. (i am being professionally vague here; i can't remember if this was nondisclosure or not.) when we actually get some delivered from sgi, i'll let you know.
hutch@fps.com (Jim Hutchison) (03/03/90)
In <43941@ames.arc.nasa.gov> lamaster@ames.arc.nasa.gov (Hugh LaMaster) wrote: >In <10538@alice.UUCP> andrew@alice.UUCP (Andrew Hume) wrote: >>[...] For example, the Sabre 2HP is only >>a little more expensive than the regular sabre, a tad smaller >>but twice as fast. >I assume that you mean 6 MB/sec. I guess Sun doesn't sell these. Are >there any problems with drivers/formatting/etc? Can you just plug them >in and run? Well, the 2-headed (2HP) version does store the defect map bit interleaved. If you are using the disk in byte or word interleaved mode, this might cause you some annoyance in getting at the defect map. >I thought the filesystem was the bottleneck. The best you can get out >of a 3 MB/sec disk is ~2MB/sec now. What kind of speeds can you get out >of a 6 MB/sec disk? Not all the loss is the file system's fault. The disk's transfer rate is not the "usable data rate". The disk handles data at 3 or 6 Mbyte/s, this "data" has sync pulses, timing gaps (gaps, pads, and delays), ECC bytes, and other little things like duplicate headers. What you generally get is about 512/600 of that (presuming your controller wants 600 byte physical length sectors and you're using 512 data byte sectors). This means you will only get about ~2.56 Mbyte/s of usable data from a 3 Mbyte/s disk or ~5.12 Mbyte/s from a 6 Mbyte/s disk. This is a lot like the difference between formatted/newfs'd disk space and raw-disk space. 600 bytes is a rough number used by SMD-controllers-I-have-known. On an IPI-2 disk/controller there are additional pads/gaps, so the number probably gets a little bigger. -- /* Jim Hutchison {dcdwest,ucbvax}!ucsd!celerity!hutch */ /* Disclaimer: I am not an official spokesman for FPS computing */