Re: [Mldonkey-users] 2.00+3

[MLdonkey]

>  I am really happy that the overnet patch 1.0 got in, but I believe that
>  telling that this patch is aggressive is a bit unfair.

It was just a comment because, on my system, it took about 5kB/s
during all the time I used it, just to publish 10 files. It was just
to say that it needed to be tuned a bit.

>  In fact, I believe
>  this patch _less_ aggressive than the previous overnet behavior... Let me
>  explain why :
>  
>  the overnet-patch1 changes the following things :
>  - fixes packets encoding/decoding, and adding new opcodes which enables us
>  to add the "overnet_publish=true" line
>  - adding a new peer storage system for the next features
>  
>  but it does not change anything related to the search/publish system
>  behavior, nor the timings/aggressivity of the overnet search system. In
>  fact, the new peer system is less aggressive as the previous one as it does
>  no more keep a list of connected peers to test, but he only _if necessary_
>  sends 16 UDP packets every 2 minutes...

Strange, as said above, on my system, it was sending  these packets
every 10 seconds. Cumulated with the searches messages, it was taking 5kB/s.

>  The real issue is that we are now adding "overnet_publish=true", it implies
>  that at boot-up (start+10s or start+60s - it won't change the bad behavior,
>  which is the originial one) mldonkey starts S*(K+1) searches, where S is the
>  nb of shared files, and K the number of keywords in file names. For 20
>  shared files and 5 keywords by filename, it implies 120 simultaneous
>  searches ! Moreover, I think that the overnet search system can be greatly
>  improved, and can be more bandwidth efficient.
>  
>  I am currently working on overnet patch2, to avoid this bandwith hog:
>  - Proggressivly publish files (1 file is published every 2 minutes)
>  - Rework search/publish system to make it more efficient (shouldn't last so
>  long in a distributed hashtable environment to find the right peers).

Yes, good ideas. We are waiting for these new features :) Maybe you
may want to join the developpers team one day for direct access to the
CVS ?

By the way, there were already errors introduced in the stats when
pango's patch was applied (in fact, it could not be applied directly,
so it was partially rewritten). It is the same for your patch (there
were some changes in the overnet code preventing direct applying your
patch), so maybe some errors were introduced.

- MLDonkey