[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [GNUnet-developers] Suggestion: CIA (The open source version control

From: Christian Grothoff
Subject: Re: [GNUnet-developers] Suggestion: CIA (The open source version control informant)
Date: Mon, 3 Sep 2007 03:03:29 -0600
User-agent: KMail/1.9.5

I'm not sure that CIA would help get a discussion going, but I agree that the 
current lack of discussion on the mailinglist(s) is not great.  Since I maybe 
to blame for some of this due to me posting news on IRC and not here, I guess 
the best way to address this maybe to just start posting more stuff here (in 
the hope that discussions will follow).

So, here's what's going on right now:

1) there is an odd crash of gnunetd in SVN HEAD that has been reproduced by a 
couple of people but cannot easily be diagnosed: the code seems to jump to a 
random address (and dies), resulting in useless stack traces, so I have no 
idea or hint as to where it is comming from -- that's the main thing holding 
off any release right now

2) the HTTP-transport using libmicrohttpd ( 
and libcurl is in alpha.  Some peers are using it, but it still seems to have 
a few quirks that need to be ironed out before it is ready for general 

3) SVN HEAD contains a new GTK interface for gnunet-setup; the new UI is much 
nicer and easier to use than the old GtkTreeView-based one, feedback would be 

4) I'm seeing quite a bit of UDP traffic (50%) going on right now.  This 
causes packets to be fragmented, and sadly quite often reassembly fails 
(which means bandwidth is wasted).  I've commited code to SVN HEAD that will 
cause gnunetd to consider switching to TCP whenever messages need to be 
fragmented.  This may help a bit.  An alternative might be to extend the UDP 
transport to try retransmitting lost fragments.  That'd cost memory and add 
complexity, but maybe worth it.  

5) Some users still seem to have problems with high CPU load; mine is less 
than 5% with similar configurations, so I it is hard for me to judge what is 
going on.  src/server/handler.c has an #ifdef that allows gnunetd to collect 
per-query-type execution timings (printed on shutdown only!), a report on a 
system running HEAD and experiencing high CPU load would be welcome. (Note: 
HEAD uses less CPU by being less persistent about re-trying session key 
exchanges with peers that are most likely down...).


Happy hacking


On Sunday 02 September 2007 14:05, donnar wrote:
> As technical details of recent developments are not discussed on this
> mailinglist: How about using CIA ( ?
> Quote: 'Do you collaborate with other developers over Internet Relay Chat?
> How often do you find yourself wondering what the other developers are
> doing right now? How many people in your project need to know immediately
> when new code is available to test or critique?'
> CIA stores the changes it receives and creates a web page and RSS feed.
> It's quite interesting for people to see how the project is advancing and
> what developer are currently working on. CIA is GPL'd btw.
> Example: CIA generated stats website of Amarok
> (
> Just a suggestion...
> Donnar
> _______________________________________________
> GNUnet-developers mailing list
> address@hidden

reply via email to

[Prev in Thread] Current Thread [Next in Thread]