alexis@panix.UUCP (Alexis Rosen) (02/14/90)
I haven't seen any resonses to Roy's posting yet, so here's mine...
From my once-broad and now hazy knowledge of AppleTalk protocols, I would asy
that this is impossible. Except-
I seem to be seeing this myself. In particular, on a dedicated AppleShare
server servicing AFP calls from multiuser databases, I've seen database records
get trashed, so that they look like full 8-bit ASCII garbage. I've seen this on
two separate networks, one LocalTalk, and one mixed LocalTalk/EtherTalk. It
seems to occur more on the mixed network (dozens of bad records instead of
three, across a three-week period). I can't believe that the database is
responsible for this. If it were writing stuff out of a bad pointer in
memory (the only way the DBMS could be responsible, I think), I wouldn't
occasionally see a dozen or so bytes or good data mixed in with the bad. It is
possible (though not confirmed) that the damaged records were the subject of
AFP locking contention.
While this doesn't match Roy's experience directly, the nature of the garbage
seems the same.
Can there actually be a bug in the AppleTalk drivers (which occurs rarely)
which fails to assemble packets correctly? I know this is _really_ farfetched,
but what else could it be??? I don't see how it could bit a bit-shift error,
since I think the SCC handles that in hardware (and certainly EtherNet does),
but that's what it seems to look like.
Mystified,
Alexis Rosen
sysadmin, Panix
alexis@panix.uucp
{apple,cmcl2}!panix!alexisalexis@panix.UUCP (Alexis Rosen) (02/16/90)
Well, somebody is sufficiently interested in this to send me mail. Whoever
you are, please post or try again. Your message was received sans body text,
but with 50 feet of mangled headers. There was no useful return address but it
seemed to be coming through a Mail*Link Qmail gateway...
Alexis Rosen
{apple,cmcl2}!panix!alexis
alexis@panix.uucpborton@fwi.uva.nl (Chris Borton) (02/23/90)
alexis@panix.UUCP (Alexis Rosen) writes: >I seem to be seeing this myself. In particular, on a dedicated AppleShare >server servicing AFP calls from multiuser databases, I've seen database record >get trashed, so that they look like full 8-bit ASCII garbage. I've seen this o >two separate networks, one LocalTalk, and one mixed LocalTalk/EtherTalk. It >seems to occur more on the mixed network (dozens of bad records instead of >three, across a three-week period). I can't believe that the database is >responsible for this. If it were writing stuff out of a bad pointer in >memory (the only way the DBMS could be responsible, I think), I wouldn't >occasionally see a dozen or so bytes or good data mixed in with the bad. It is >possible (though not confirmed) that the damaged records were the subject of >AFP locking contention. This might possibly be related to the problem I reported months ago with MPW Projector over EtherTalk. We were having problems with source files not coming back the same way they went in, and doing it twice in a row did not result in the same file. This was occurring only over EtherTalk; LocalTalk didn't do this at all. It also *seemed* to make a difference which volume we used. All very strange stuff with way too many variables to trace down in a sane fashion. The files would generally be almost whole, except for 2-5 lines either missing (between random character locations) or garbage. I received a note from a fellow at Kinetics when I posted about it saying that he had seen some strange stuff like this too, but was simialarly mystified. The problem has been reported to Apple a couple of times but we have received no repsponse on it. -cbb Chris Borton borton@fwi.uva.nl Mac Developer & AppleTalk Network Administrator, University of Amsterdam CS