home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.sys.sun.hardware:5738 comp.periphs.scsi:5543
- Newsgroups: comp.sys.sun.hardware,comp.periphs.scsi
- Path: sparky!uunet!wupost!gumby!destroyer!ncar!csn!teal.csn.org!milton
- From: milton@teal.csn.org (Milton Scritsmier)
- Subject: Re: Problems with Fast SCSI-II on Sun Sparc 10 (long)
- Message-ID: <Bxy9x9.MEH@csn.org>
- Sender: news@csn.org (news)
- Nntp-Posting-Host: teal.csn.org
- Organization: Colorado SuperNet, Inc.
- References: <MRD.92Nov13160506@morsel.stdavids.picker.com> <1992Nov15.121008.12717@pt.com> <37088@cbmvax.commodore.com>
- Date: Thu, 19 Nov 1992 06:27:07 GMT
- Lines: 40
-
- In article <37088@cbmvax.commodore.com> jesup@cbmvax.commodore.com (Randell Jesup) writes:
- >jjg@pt.com (John Grana) writes:
- >>As one who has spent the past year dealing with Fast SCSI on single ended
- >>busses, all on SBus, I can offer the following:
- >
- >> Also, our driver performs a speed
- >>reduction when it detects a sync. problem. So, the system would start at
- >>10 MB/sec., then drop to 8 and maybe even 6.7 MB/sec before it would
- >>stabilize.
- >
- > What method do you use to determine when a sync problem has occurred?
- >Mismatched transfer sizes? How do you tell this is different from either
- >commands which return less than requested, or people who use CAM (or the
- >equivalent) and tell you the command will do more or less IO than it actually
- >does?
- >
-
- Your're right, that's a real problem. While there are cases where you
- definitely know a sync problem has occurred (eg., offset count
- exceeded), there are also ambiguous cases where you don't know for
- sure (eg., the transfer hangs).
-
- The best way I can think of uses the assumption that sync problems
- are chip problems while transfer count problems are due to firmware
- problems (including programming the chip with the wrong count).
- Thus if you renegotiate for a slower synchronous transfer speed (or
- better yet for an asynchronous transfer), retry the command that
- failed, and have it succeed, then probably you are dealing with a
- sync problem. This assumes the firmware on the target goes through
- the same basic steps in either case, except for the setting of the
- transfer mode and speed.
-
- There could be firmware bugs which are timing dependent and hence
- sensitive to the transfer rate, or firmware bugs where the chip
- is programmed wrong for certain transfer modes. However, from the
- host's point of view, it probably doesn't matter. With one transfer
- rate the command fails, while with another transfer rate the command
- succeeds. Whatever the cause, the work-around is the same.
-
-
-