Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uwm.edu!uakari.primate.wisc.edu!brutus.cs.uiuc.edu!psuvax1!psuvm!CRNLNS!DSR From: DSR@CRNLNS.BITNET Newsgroups: bit.listserv.i-amiga Subject: RE: Starboard SCSI Module Message-ID: Date: 11 Jan 90 19:05:00 GMT Sender: Info-Amiga List Reply-To: Info-Amiga List Lines: 47 Approved: NETNEWS@PSUVM Gateway X-Original-To: i-amiga@ubvm.BITNET Comment: forwarded by CRNLNS/FMAIL v2.0 Comment: REPLY may not work. Comment: Network-Source: _LNS62::DSR (HEPnet/SPAN) Comment: Originally-From: DSR "Dan Riley" Comment: Originally-To: JNET%"i-amiga@ubvm" [apologies for any bogus messages I sent...we just upgraded to VMS 5.3, and the new DECwindows clients don't seem to like the MIT X11R4 server on my DECstation, so weird things have been happening] MARK GOODERUM - UNIV. OF KANSAS ACS - MARKV@UKANVAX writes: >For the Mountlist, leave the first two cylinders unused (0 and 1) for >the new driver (don't have it, just heard about it) which (I think) >supports the 'hardblocks' standard. The driver with Rigid Disk Blocks support doesn't exist yet--it's still being worked on. But do reserve a couple of cylinders. >Second, 'physically' format the drive with a 1:1 interleave, and leave it up >to DOS to take care of the interleave. (With the Interleave= entry in the >Mountlist.) Depending on your system your want a 2:1 or 3:1 interleave. This goes directly against my experience, my common sense, and Microbotics' recommendations. I've gotten the best performance using a DOS interleave of 0 (in the MountList), and hard formatting the drive with an interleave of 2:1 or 3:1 (which is better depends on the drive you are using). Using the DOS interleave adds lots of transfer overhead--with a DOS interleave of 0, FFS can request large contiguous reads whenever possible. With a non-zero DOS interleave, FFS has to send out a seperate request for every block, which adds overhead and restricts any optimizations the driver may make for large reads. With OFS, this makes little difference, but with the FFS it matters, since FFS is lots smarter. >On interleave I've figured this out (at least the StarDrive.) Mark's description is basically correct. I just wanted to note that there is an update to the StarDrive software which, among other improvements, does correctly explain the difference between DOS and SCSI interleaves, and gives some timings and recommendations. It also explains how to set up a MountList entry, and includes support for Adaptec adapters and fixes some bugs in the driver. I pulled my copy off the Microbotics support conference on BIX, but the docs specifically disallow any redistribution. Hopefully new StarDrives should come with the update--if not, call Microbotics and ask for it. -Dan Riley (dsr@crnlns) -Wilson Lab, Cornell University