[M3devel] next problem (NT386GNU)

Mika Nystrom mika at async.caltech.edu
Wed Jan 23 06:58:55 CET 2008


I think Windows 2000 is well known for having some exceptionally
nasty bugs in its TCP implementation.  In fact my Modula-3 code has
special workarounds for win2k: it doesn't download more than a few
kilobytes at a time---if you try more, the network card sometimes
goes catatonic.  There are some notes on it here and there on the
Internet.  Not sure what the cause is.

    Mika 

Olaf Wagner writes:
>Quoting Randy Coleburn <rcoleburn at scires.com>:
>
>> Jay / Olaf:
>>
>> I installed CVSNT and TortoiseCVS.  I am having some trouble with
>> "update" however.  When I try to update the whole tree, it begins
>> working, but after a little while I start seeing messages that checksums
>> didn't match on certain files and that it will refetch.  Then, after
>> just a little longer, it stops responding.  I've left it running for
>> hours, but no new lines scroll by saying it is doing anything.  Finally,
>> I just hit cancel and it promptly stops and informs me it aborted the
>> operation, as if it been waiting all along for me to tell it to give
>> up.
>>
>> Now, if I update individual files or selected files, it seems to work
>> no problem.  And, I've been able to check out the whole repository, and
>> yes it does take some time to pull everything down.  I can also do diffs
>> on selected files.  I've also used it to commit a few files to the
>> repository.
>>
>> So, it is working fine, except for update.  Either something is amiss,
>> or I'm not doing something just right.  I am about to consider going
>> back to cygwin and running its version of cvs on my tree to see if it
>> will do the update, but I seem to recall reading somewhere that is not
>> good to mix between cvs and cvsnt on the same tree.
>
>This is strange. I've never seen that myself before. A complete
>update or checkout should take several minutes via encrypted
>TCP/IP, but not longer (depending on the line bandwidth and latency,
>of course).
>
>Just today I've heard of a customer who suffers from random connection
>problems with Subversion and Windows 2000 workstations; the theory there
>being that some TCP/IP parameters are set very strange and the whole
>protocol stack is not up to the task in its default configuration.
>This does not happen on XT or Vista, though, who seem to have incorporated
>significant improvements. (It's all rather strange to me, since I'm a
>Unix person myself and used to stable networking.)
>
>So depending on your OS you might want to tweak the TCP parameters a
>little bit to see if it improves. Hard-core debugging of this problem
>will need careful use of tcpdump and produce megabytes of logs;
>but perhaps just using the cvs trace option (cvs -t up ...) will
>give some indication if there's another problem than the underlying
>connection.
>
>There's also nothing against checking out a different workspace
>with the Cygwin cvs to see if that works better. We've made good
>experiences with CVSNT though.
>
>If the problem persists, we could also try to do some tcpdumping on
>the server side (Ronny Forberger would be the one to do it), but this
>will take some preparation time.
>
>@Ronny, perhaps you've got another idea?
>
>Regards,
>
>Olaf
>-- 
>Olaf Wagner -- elego Software Solutions GmbH
>                Gustav-Meyer-Allee 25 / Gebäude 12, 13355 Berlin, Germany
>phone: +49 30 23 45 86 96  mobile: +49 177 2345 869  fax: +49 30 23 45 86 95
>    http://www.elegosoft.com | Geschäftsführer: Olaf Wagner | Sitz: Berlin
>Handelregister: Amtsgericht Charlottenburg HRB 77719 | USt-IdNr: DE163214194



More information about the M3devel mailing list