home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!sgigate!odin!twilight!zuni!anchor!olson
- From: olson@anchor.esd.sgi.com (Dave Olson)
- Newsgroups: comp.sys.sgi.bugs
- Subject: Re: Fix for 4.0.1 / 8M RAM slow disk writes
- Message-ID: <u2hmcuc@zuni.esd.sgi.com>
- Date: 26 Dec 92 02:44:40 GMT
- References: <1992Dec24.013446.26848@donau.et.tudelft.nl>
- Sender: news@zuni.esd.sgi.com (Net News)
- Organization: Silicon Graphics, Inc. Mountain View, CA
- Lines: 48
-
- In <1992Dec24.013446.26848@donau.et.tudelft.nl> reinoud@dutecai.et.tudelft.nl (R. Lamberts) writes:
-
-
- | OK guys, don't laugh. I still have 8 meg and IRIX 4.0.1,
- | and was getting sick of the slow disk write bug. If you
- | are in the same situation (8 meg / 4.0.1), this crazy
- | fix may be of interest to you.
- |
- | It appears that doing a file system sync 20 times a second
- | increases disk write speed enormously, without giving too
- | much overhead. So compile something like this:
- |
- | main()
- | {
- | do
- | {
- | sync();
- | sginap(5);
- | }
- | while (1);
- | }
- |
- | (Please don't whine about this not being the best example
- | of software engineering. ;-)
- |
- | If you use applications that write to disk at very high
- | rates, you may try to decrease the sginap argument, but
- | that results in a noticeable system load. Most tools
- | run at maximum write speed with the code shown above.
- |
- | Run this thingy in the background and you'll actually start
- | enjoying large cp or file un-tar actions...
-
- The best fix is to get 4.0.5. This bug was introduced in the last days
- of 4.0.1, and fixed in 4.0.2 (after I found it (I wasn't the one who
- introduced this one!) and all later releases. The code has changed
- enough since then that I don't want to invest the time to see just
- what else might help; certainly dropping BDFLUSHR to 1 would have
- similar effects, without using up another process slot and causing
- more context switches. I seem to recall that increasing NBUF would help
- prevent the situation from arising, but I may be remembering wrong.
-
- Doing it ever 50 msecs (as your program does) seems just a *bit*
- excessive though!
- --
- Let no one tell me that silence gives consent, | Dave Olson
- because whoever is silent dissents. | Silicon Graphics, Inc.
- Maria Isabel Barreno | olson@sgi.com
-