Re: [Help-gnunet] OK to use a FQDN for 'IP' in gnunet.conf?

[Ludovic Courtès]

Hi again!

One day, 21 hours, 50 minutes, 20 seconds ago, 
Christian Grothoff wrote:
> No, we pretty much use hardly any non-blocking calls for most of our IO 
> (there 
> are some exceptions for sockets, but they boil down to read/write calls that 
> center around a select or error handling) and pretty much all selects are 
> fully blocking.  This is one reason why we have so many threads ;-).

>From what I can see with `grep PTHREAD_CREATE *.c' (in src/server),
there should be 4 threads: the main thread which does nothing but wait
for a signal, a TCP listener thread (the one calling `select ()'), two
packet processing threads (created by `initCore ()'), and a "cron"
thread (created in util/cron.c).  The select-thread should spend most of
its time waiting for messages.  The two processing threads may as well
spend a fair amount of time waiting for new packets.  The cron thread
sleeps at most 2 seconds in between two "jobs" and there seems to be a
significant number of jobs.  I believe this thread may create a certain
level of almost permanent CPU activity; combined to the processing of
incoming requests, this may be what yields to 100% of CPU usage at
times.  In such a scenario, the current decisions made depending on the
value returned by `getCPULoad ()' can't help honor the CPU quota.  Does
this make sense?

Note that `ps m' shows that my running `gnunetd' has 8 threads so I
guess I'm still missing a few threads.  :-)

Do other people feel like gnunetd is eating too much CPU time or am I
the only one noticing this?

Thanks,
Ludovic.