[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [GNUnet-developers] bugs. this and that.
From: |
Christian Grothoff |
Subject: |
Re: [GNUnet-developers] bugs. this and that. |
Date: |
Fri, 21 Jun 2002 20:55:35 -0500 |
User-agent: |
KMail/1.4.1 |
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On Friday 21 June 2002 08:54 am, you wrote:
> From: Igor Wronsky <address@hidden>
>
> > did get >1MB of 4MB before gnunetd went postal. I'll compile
> > gnunetd with gprof and try to reproduce the effect and
> > find the load source.
>
> Just do a gdb attach and look it see where it is at.
> Usually it is in a tight loop somewhere...
Well, good idea, but didn't seem to apply in this case. So far, it looks like
my initial finding could have been the real issue (since I can't reproduce
the problem anymore). The problem was that in our database (not the gdbm db)
we were handling collisions in the hashtable by falling back on a linear
scan. When that collision list got to long, it became possible that we were
not done with one scan when the next was already started -- giving the
illusion of 'never finishing' where we were really doing different (stupid,
wrong, inefficient) work all the time. A more elegant solution for the
database is in the CVS now, and I haven't seen this problem since.
Christian
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: For info see http://www.gnupg.org
iD8DBQE9E9kX9tNtMeXQLkIRAh6TAJ9PkiUKUOsyil6yffUSPxdkZNDECgCfY+iu
5GQ5eBb7jFMOXwBP9pVl51E=
=5Pxx
-----END PGP SIGNATURE-----
Re: [GNUnet-developers] bugs. this and that., Wayne Scott, 2002/06/21