Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!mnetor!seismo!rutgers!princeton!allegra!ulysses!gamma!mb2c!edsdrd!edstb!msudoc!haas From: haas@msudoc.UUCP Newsgroups: comp.unix.questions Subject: Re: Cheap SCSI disks for Cheap Sun's? Message-ID: <1421@msudoc.UUCP> Date: Tue, 5-May-87 05:34:39 EDT Article-I.D.: msudoc.1421 Posted: Tue May 5 05:34:39 1987 Date-Received: Thu, 7-May-87 07:20:19 EDT References: <106500005@datacube> Reply-To: haas@msudoc.UUCP (Paul R Haas) Organization: Michigan State Univ., Engineering, E. Lansing Lines: 68 Summary: report on Maxtor XT-3280. The Sun that I am using has a 240 MByte (280 MByte unformatted) Maxtor XT-3280 on it. We bought it from Delta Microsystems. They include the appropriate cable and a formatting program so that you can plug it in to a 3/50 and run. It does not include a tape drive, so you have to configure the system with another system. I have never tried to run my machine independently of the rest of the network. I still boot off a diskfull server with an eagle. In general the system seems the same or a little faster with a local Maxtor as compared to a remote eagle. There are some benchmarks later in this posting with numbers. Here are the particulars: The drive costs about $4200. It is sold by: Delta Microsystems Rt. 1 Box 129 C Brentwood, CA 94513 (415) 833-2980 It is about 10 inches wide by 6 inches high by 13 inches deep. The shipping weight was 21 lbs. When formatted there are 475800 sectors available to Unix (237.9 MBytes). The average access time is 30 msec. I ran the benchmarks from: Kridle, R., McKusick K., Performance Effects of Disk Subsystem Choices for Vax Systems Running 4.2BSD UNIX, In USENIX Summer 83 Conference Proceedings, July 1983, pp. 155-169. They sequentially read or write an 8MByte file with 4k or 8k blocks. Here are the results of the Csh time command for each benchmark. I ran each test three times and took the best one, on a mostly idle system. Sun 3/50 with Maxtor drive: write_8192 0.0u 8.6s 1:02 13% 0+1k 14+1785io 0pf+0w read_8192 0.0u 5.1s 0:31 16% 0+1k 1026+1io 0pf+0w rewrite_8192 0.0u 6.4s 1:17 8% 0+1k 492+1025io 0pf+0w write_4096 0.0u 9.6s 1:00 15% 0+0k 11+1701io 0pf+0w Sun 3/50 to 3/180 nfs server with Fujitsu Eagle: write_8192 0.0u 7.1s 3:15 3% 0+1k 1+1025io 0pf+0w read_8192 0.0u 5.9s 0:32 18% 0+1k 1028+1io 0pf+0w rewrite_8192 0.0u 10.0s 1:44 9% 0+1k 3+1025io 0pf+0w write_4096 0.2u 9.6s 2:51 5% 0+0k 1027+1025io 0pf+0w Sun 3/180 with Fujitsu Eagle: write_8192 0.0u 6.1s 0:13 44% 0+1k 11+1048io 0pf+0w read_8192 0.1u 3.3s 0:21 16% 0+1k 1027+1io 0pf+0w rewrite_8192 0.0u 3.7s 0:15 24% 0+1k 25+1025io 0pf+0w write_4096 0.1u 7.3s 0:14 51% 0+0k 10+1043io 0pf+0w On a real program which was bound by paging speed (it was only reading memory not writing it), it ran about 20% faster with the local disk than over the ethernet. I am willing to run benchmarks that are mailed to me. I can mail the benchmark programs to anyone who is interested. They are quite small. At some point we will go to the effort of hooking up a couple diskless 3/50's to the diskfull one and see how it behaves as a server. Judging by the elapsed time, I would guess it would only be good for two or maybe three clients. --- disclaimer: I am only a customer of Delta Microsystems, however the paperwork is still being processed, so I don't want them to get too mad and take back their disk before it becomes our disk. --- Thanks to Barry Shein bzs@bu-cs for supplying the benchmarks. --- Paul Haas haas@msudoc.egr.mich-state.edu (35.8.8.108) ...!ihnp4!msudoc!haas