marantz@aramis.rutgers.edu (Roy Marantz) (12/25/87)
I'm looking for a recomendation for a smallish (20 to 40 Mb would be
idea) SCSI disk that I can attach to a SUN-3/50. I want to use the
disk for swapping (and maybe /tmp). I'm hoping this will speed up
(over nd swapping) systems that are using alot of swap space and/or
trashing. The 3/50 is fairly prone to thrashing due to it's small
(4Mb) real memory size. Any comment or suggestions to where to look
would be appreciated.
Roy
uucp: {most places}!rutgers.edu!aramis.rutgers.edu!marantz
arpa: marantz@aramis.rutger.edu
--
uucp: {ames, cbosgd, harvard, moss}!aramis.rutgers.edu!marantz
arpa: marantz@aramis.rutger.eduroy@phri.UUCP (Roy Smith) (12/28/87)
In <2604@aramis.rutgers.edu> marantz@aramis.rutgers.edu (Roy Marantz) writes: > [...] SCSI disk that I can attach to a SUN-3/50. I want to use the > disk for swapping (and maybe /tmp). I'm hoping this will speed up > (over nd swapping) At the time we were looking at Suns (about 2 years ago; the 3/50's had just been announced), we were thinking that lots of small disks on the clients would be faster than sharing a big disk on a server for exactly the reasons stating in Roy Marantz's article. We were slavishly following the Unix truism that "performance is directly proportional to the number of independant disk arms". After some discussions with the folks at sun, we were convinced that we had the wrong idea. Sun claims (and I have no reason to doubt them) that you get better disk throughput with a remote Eagle than with a local SCSI disk. In other words, the performance hit you take for going through the network is less than the difference in speeds between the two types of drives. I recently saw somebody (on Sun-Spots, I think) talking about his 2-disk server; he wanted to put ND partitions on both disks and have each client mount two swap partions; one on each disk. I suspect that idea will win a lot more than buying a SCSI disk to swap on. I'm very tempted to try it myself. This was two years ago. I havn't really been paying much attention to SCSI disks in that interval, so it is possible that they have gotten fast enough to make the comparison above no longer true. On the other hand, SMD disks have gotten faster also (a Super Eagle on one of the new fast VME controllers should be significantly faster than a plain Eagle on a XY-450). One thing I know has changed. At the time, the cost per Mbyte curve for disks made the bigger disks a lot more attractive compared to the 20-40-70 Mbyte SCSI disks. I think small disks have dropped in price faster than big ones over the past couple years, making the cost per Mbyte much more even accross the spectrum of disk capacities. -- Roy Smith, {allegra,cmcl2,philabs}!phri!roy System Administrator, Public Health Research Institute 455 First Avenue, New York, NY 10016
mangler@cit-vax.Caltech.Edu (Don Speck) (01/15/88)
In article <3091@phri.UUCP>, roy@phri.UUCP (Roy Smith) writes: > I think small disks have dropped in price > faster than big ones over the past couple years, making the cost per Mbyte > much more even accross the spectrum of disk capacities. What has mostly happened is that small disks have grown in capacity without getting much more expensive, and the big disks have mostly stood still. These days, anything over 9 inches doesn't make sense, because larger-diameter disks don't offer any higher capacity or better price ratio. The 8-inch disks are rapidly catching up (I know of at least six 8-inch drives with at least SuperEagle capacity). The reason, I think, is that the larger disks are pushing against the limits of SMD controllers, not the limits of the media. The CDC 9771 14-inch drive is a good example; it's built with 1064 cylinders, but on typical controllers you can only use 1024, and they had to grossly slow down the rotation rate in order not to exceed the bit-rate limit of many controllers. And that's an old drive. Perhaps IPI offers some hope (nobody is going to take SCSI seriously for big drives), but it may be too late. 8-inch drives are pushing the controllers just as hard. CDC's latest has 1635 cylinders and transfers at 24 MHz. In a couple of years, we'll be seeing 5.25 inch drives with the same specs. What are we going to hook them up to? Not ESDI, it has an even lower bit rate than SMD. The only option left will be native SCSI, and it will need lots of buffering, because even the Sun-3/60 SCSI bus appears to be limited to 0.8 MB/s. Don Speck speck@vlsi.caltech.edu {amdahl,ames!elroy}!cit-vax!speck