home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.os.linux
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!uakari.primate.wisc.edu!ames!agate!boulder!juliet!drew
- From: drew@juliet.cs.colorado.edu (Drew Eckhardt)
- Subject: Re: Slow SCSI performance
- Message-ID: <1992Aug31.194713.8233@colorado.edu>
- Sender: news@colorado.edu (The Daily Planet)
- Nntp-Posting-Host: juliet.cs.colorado.edu
- Organization: University of Colorado at Boulder
- References: <1992Aug31.162211.8065@zip.eecs.umich.edu>
- Date: Mon, 31 Aug 1992 19:47:13 GMT
- Lines: 49
-
- In article <1992Aug31.162211.8065@zip.eecs.umich.edu> yhe@quip.eecs.umich.edu (Youda He) writes:
- >I had this problem from 0.97pl1, and now pl2, when SCSI is accessed, the
- >system seems stop to respond to other vc, and SCSI itself is SLOW. sometimes
- >the hard drive LED light but nothing happened, when this happened, I usually
- >can issue some command on other vc and the "deadlock" on SCSI seemed resolved.
- >the system is:
-
- 1. SCSI transfer rates should be slow, 50-100k/sec on systems with unbuffered
- disks, (~300k/sec on buffered drives) or disks that require mode page
- twiddling to enable the buffers.
-
- This will be fixed in R2 of the SCSI drivers - basically, a design
- flaw results in the routines reading 1 block per revolution -
- on a 3600 RPM drive, this comes to ~60k/sec.
-
- 2. Accessing SCSI disks on one tty / pty /ttys / etc should not
- affect performance of a non-I/O bound process on another
- tty. It doesn't on my system.
-
- A lot depends on the specific low level driver.
-
-
- >33 MHZ 486 with AHA1542B Maxter 8760S, the files systems are sda3 32MB swap
- >(I don't know I need it since I have 16MB memary), sda4 64MB minix, sda5 100MB
- >on logical device, sda5 200MB also a logical device. sda[56] are extended files
- >systems. I guess the problems may be that I use extended file systems on
- >extended partition on a SCSI drive? can I repartition the system that not use
- >the extended partition solve the problem? The other partitions are used for
- >OS/2, so the system must coexist with OS/2.
- >
-
- Extended / "Normal" partitions don't make a difference as far as the driver
- is concerned. As far as the driver is concerned, the only difference
- between extended and normal partitions is during initialization -
- when you actually access them, they're just a starting sector
- and a limit to the device driver.
-
- The only problem you might see comes from the Linux block device code -
- requests are sorted by major, minor, and block number. With different
- partitions, requests may be "out of order" with respect to the disk
- blocks - which will degrade performance as the thing has to seek.
-
-
-
- --
- Microsoft is responsible for propogating the evils it calls DOS and Windows,
- IBM for AIX (appropriately called Aches by those having to administer it), but
- marketing's sins don't come close to those of legal departments.
- Boycott AT&T for their absurd anti-BSDI lawsuit.
-