Path: utzoo!mnetor!tmsoft!torsqnt!news-server.csri.toronto.edu!cs.utexas.edu!usc!apple!agate!eris.berkeley.edu!korn
From: korn@eris.berkeley.edu (Peter "Arrgh" Korn)
Newsgroups: comp.protocols.appletalk
Subject: Re: Problems with Nodems /  Try Compatible Systems
Message-ID: <1990Dec24.085714.15847@agate.berkeley.edu>
Date: 24 Dec 90 08:57:14 GMT
References: <1990Dec19.033827.23551@monu6.cc.monash.edu.au> <2472@sparko.gwu.edu> <1990Dec21.144218.17923@engin.umich.edu>
Sender: usenet@agate.berkeley.edu (USENET Administrator)
Organization: What, me organized???
Lines: 41

In <1990Dec21.144218.17923@engin.umich.edu>, John_Mansfield@emal.sprl.umich.edu (John F. Mansfield) said:  

>I have had no experience with the Nodem, however one of the publications 
>did a review of Nodems, Ether+ and the NuvolinkSC. They (MacWorld or MacWeek)
> said that there wasnt much in it, but they would choose the Ether+ over the 
>Nodem and the Nuvolink was the same animal as the Ether+ in a different box.
....
>There is a problem with these SCSI ethernet devices and CE Software's Quickmail
>at some point, if you use the SCSI ehternet box on the MailServer, both the 
>SCSI box and the QM Server try and grab the same interupt and the Mac hangs.  
>This explanation was offered by CE and they do not have a work around.  

There was a problem with the low level SCSI manager in early System 6.0.x
versions of the OS.  It's likely still there, though I understand it's to
be fixed in 7.0.  The problem:  it's not re-entrant.  And in certain, very
very rare cases, with a heavily stressed AppleShare server multiple AFP
requests simultaneously over a SCSI ethernet connection to/from a SCSI
disk, the mac would hang; typically somewhere deep in HFS ROM code.

It wouldn't surprise me if you are seeing this problem with Quickmail.  A
heavily used e-mail server, if it schedules it's multiple e-mail accesses
in the wrong way, could hit this problem.

For this reason Apple doesn't recommend using SCSI-Ethernet connections on
AppleShare servers.  

Oh, as to what "very very rare" means:  5 other macs on ethernet doing
continuous file & directory copies/creates/deletes simultaneously would
usually trigger this problem after several days.  Figuring 1 file/directory
being created/deleted each second on each mac, that works out to it
crashing after simultaneous stressing of 1,620,000 transactions.  Pretty
rare.  It didn't help that the server was a Mac Plus (vs. an SE -- the Plus
has the least impressive SCSI implementation...).


Peter
 -- Moonlighting at Nuvotech when I'm not busy reading netnews...
--
Peter "Arrgh" Korn
korn@mica.Berkeley.EDU
{decvax,hplabs,sdcsvax,ulysses,usenix}!ucbvax!mica!korn