Re: CVS slow and TCP Compression

[Daniel Beckham]

Someone else might be able to answer the question as to whether it has
anything to do with cvs, but I would think it's a simple matter of network
performance.   It would make sense that using compression will speed up the
checkout process, but what kind of speed problems are we talking about?  Do
you mean, a 2MB file take 30 minutes to download or 1 minute?  And how much
faster does a 2MB file checkout with compression on?  What sort of network
are you running?  When you say tons of collisions, what do you mean?
Collisions on an Ethernet segment is a normal thing.  If I remember
correctly, under 3% is ok, but more than that could indicate a network
issue.

Daniel

----- Original Message -----
From: "Scott Willy" <address@hidden>
To: <address@hidden>
Sent: Sunday, July 01, 2001 5:39 AM
Subject: CVS slow and TCP Compression


> We have installed cvs on a Cobalt Cube (Linux) and had some bizarre
behavior
> which we would like to share.
>
> We have been using cvs on a Solaris server for 18 months or so and decided
> to get a Cube and dedicate it to cvs. After installing cvs on the cube we
> did some tests and all seems to be working fine. We took repositories with
> 1000's of files of source code and do checkouts no problem (and very
fast).
> Then...
>
> We did a test with repositories containing libraries of big (100KB-2MB)
> jar/dll/exe files (and not C++ or java source code). After a few files,
the
> checkout would slow to a craw. If we left it long it enough it would
finish
> the checkout, but it was too slow to be usable.
>
> We found a workaround by using the -z option to set the TCP compression to
> something other than zero. Anything between 1-9 seems to work just fine.
The
> checkouts of the large files is fast.
>
> Bizarre...
>
> Another observacation we had, was that when doing a checkout we would get
> tons of Ethernet collisions on the LAN. This was true for both the Solaris
> server we have in production and the Cube we were testing. This was true
> even if we only had cvs server and one client on the LAN.
>
> We still had tons of collisions using our workaround (-z), so they might
not
> be part of our problem, but...
>
> We would be most interested if anybody else has seen something like this
or
> has any comments.
>
>
> _______________________________________________
> Info-cvs mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/info-cvs
>