home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!snorkelwacker.mit.edu!ai-lab!CATFISH.LCS.MIT.EDU!metcalf
- From: metcalf@CATFISH.LCS.MIT.EDU (Chris Metcalf)
- Newsgroups: comp.unix.ultrix
- Subject: SUMMARY: disk I/O slowdown with Ultrix 4.3
- Date: 9 Nov 1992 15:02:32 GMT
- Organization: MIT Laboratory for Computer Science
- Lines: 40
- Distribution: world
- Message-ID: <1dlui8INN805@life.ai.mit.edu>
- References: <1992Oct26.181727.8322@mintaka.lcs.mit.edu>
- NNTP-Posting-Host: catfish.ai.mit.edu
-
- A week ago I wrote:
-
- > Why do I *lose* performance with maxcontig = 8 in some configurations?
-
- The configuration in question was the RZ56. Following up on a suggestion
- from Marcus Ranum, I tried testing the disk using "time dd" and various
- different blocksizes on the raw disk devices. As you would expect,
- performance improved significantly when using blocks larger than 8K ...
- _except_ when using blocksizes of 64K to about 80K. For some reason,
- on the RZ56 blocks in this range perform noticeably worse than 8K blocks.
-
- Dropping maxcontig on the RZ56s back to 7 yielded a dramatic improvement:
- about 35% faster on reads than maxcontig of 8. Below are the results I
- got for filesystem I/O when changing maxcontig from 1 to 8 (or 7 for RZ56s):
-
- System Disk Read Write
- DS 5000/240 RZ56 x1.15 x2.70
- DS 5000/240 RZ55 x1.00 x2.25
- DS 3100 RZ55 x1.30 x2.40
-
- I believe swap I/O has not changed, since the value of "swapfrag" is 64
- on both Ultrix 4.2 and 4.3 systems I looked at, and it has been suggested
- (thanks again mjr) that swapfrag is the unit of swap allocation and I/O.
-
- > 2. Why does the CAM subset noticeably decrease performance on some tests?
- > 3. Why can't I mount my CDROM after I install CAM?
- > 4. What are all those annoying .proto.* and .new.* files for?
-
- I never got any answers to these questions, although Nicholas de
- Metz-Noblat <nmn@cernvm.cern.ch> sent me some C code to eject CDROMs.
- I have decided not to install CAM, since I don't plan on adding devices
- to my SCSI chains anyway, and everything works just fine with the default
- 4.3 drivers. I also went ahead and deleted all the .(proto|new)* files,
- and haven't had any difficulties as a result.
-
- Many thanks again to Marcus Ranum for a helpful email dialogue.
- Having overcome the RZ56 problem I am very happy with Ultrix 4.3 now.
- --
- Chris Metcalf, MIT Laboratory for Computer Science
- metcalf@cag.lcs.mit.edu // +1 (617) 253-7766
-