home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.os.coherent
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!news.yale.edu!lasagne.med.yale.edu!silver
- From: silver@lasagne.med.yale.edu (Scot Silverstein MD)
- Subject: Disk I/O speed "problem" ... NOT?
- Message-ID: <930125092946@lasagne.med.yale.edu>
- Nntp-Software: PC/TCP NNTP
- Keywords: hard disk
- Lines: 30
- Sender: news@news.yale.edu (USENET News System)
- Nntp-Posting-Host: lasagne.med.yale.edu
- Reply-To: silver@lasagne.med.yale.edu
- Organization: Yale University
- Date: Mon, 25 Jan 1993 09:29:46
-
- I've done some experimentation. It seems the "slow" disk I/O
- on my 386sx/16 toy home machine with Seagate ST1102 89 meg IDE drive
- is not so "slow" after all under Coherent 4.0.1.
-
- After setting the m and n values to 234, 9 and running iozone
- at a 2-meg test size, I get I/O speeds of about 79k/sec for both
- reading and writing, in multiuser mode. Yet, under MSDOS with
- CORETEST.EXE, it says about 1000k/sec. The discrepancy was puzzling.
-
- Until I realized CORETEST uses a 64K "transfer block size" as a default
- and iozone uses 0.5k (512 byte block size) which is 'native' to the OS.
-
- When I run CORETEST using a 1K "transfer block size" (the minimum), lo
- and behold, it says the throughput is about 40k/sec or something like that!
- In fact, I get the "best" results from CORETEST when the "block size"
- is 37k or so on my machine.
-
- QAplus diagnostics on the hard drive say about 40K/sec also, using
- a 1k bloxk size also, I believe. The machine is at home and I didn't
- write that figure down so it may be incorrect, but it's in the ballpark.
-
- MSDOS accesses the disk in bigger chunks, apparently. That's why
- a 300,000 byte .exe file loads in a second or so. So, COHERENT
- is making efficient use of the hardware considering the nature of its
- interaction with Mr. Hard Drive.
-
- Hopefully there is no big error in facts or logic in the above.
-
- Comments?
-
-
-