Re: [Help-gnunet] two problems

[Christian Grothoff]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Saturday 04 October 2003 08:28 pm, Tracy R Reed wrote:
> After a long time away I have decided to try GNUnet again. I am running
> Debian unstable and ran into a problem getting it to compile. The testrsa
> check would fail and the node would segfault when ran. I strace'd the
> testrsa progam and it seemed to be getting hung up trying to talk to
> /dev/random. However I noticed that there were deb's available for the
> latest gnunet so I installed that and it worked just fine.

There was a problem with 0.5.5, gcry and a fresh peer setup.  Should be fixed 
in 0.6.0 and Glenn patched the debian binaries.

> So now I have working gnunet binaries. But I cannot seem to stay connected
> to any other GNUnet hosts. Sometimes I have around 10 or 20 connections
> but 90% of the time I have no peers connected. I get lots of messages in
> the logs like this:
>
> Oct  4 18:12:51 WARNING: CRC failed from
> 6B6615255C1BFED189141EC50516B286AEE71FAC, msg ignored (wrong skey?)
>
> and
>
> Oct  4 08:34:14 WARNING: makeSesionKeySigned: cannot encrypt sessionkey,
> other host not known!
>
> I did copy a bunch of hosts from http://www.ovmj.org/GNUnet/hosts/ into
> the data/hosts directory to get things started so perhaps that has
> something to do with it.

This just screams "CPU load" to me.  Remember, if you get your CPU super-busy 
(with *other* applications), GNUnet will scale back its CPU use.  So if 
you're compiling a Linux kernel, you're likely to end up with 0 connections 
since GNUnet dropped all messages for a while.  It will rebound once the load 
passes.  And since it may have dropped a couple of sessionkeys, it is likely 
to get a couple of CRC errors in the meantime.  

The load monitoring has been improved in 0.6.0, but nevertheless this can 
still occur if you configure GNUnet to any CPU level that is smaller than 
100% and if other applications drive the utilization to a higher level.  
Anyway, the problem should be much less dramatic in 0.6.0 since the CPU 
utilization leads to message discards in fewer cases (CPU load is dominated 
by the knapsack solver, so avoiding processing in other places makes little 
sense anyway).

Christian
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.7 (GNU/Linux)

iD8DBQE/ga2C9tNtMeXQLkIRAqusAJ9TCZjS8D+akZRX2axDWL6KbDyTbwCeOqrD
0bjI9wFVxcm+cnMZ7Vu0ODk=
=Sx/h
-----END PGP SIGNATURE-----