mldonkey-users
[Top][All Lists]
Advanced

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

Re: [Mldonkey-users] Accepted/rejected pango patches


From: MLdonkey
Subject: Re: [Mldonkey-users] Accepted/rejected pango patches
Date: Wed, 11 Dec 2002 11:43:20 +0100

>  > NOT APPLIED, need some discussion:
>  >      17_hide_chunks_being_uploaded: 
>  >      21_fifo_new_sources:
>  >      25_bolder_remove_old_sources:
>  
>  Admiteddly #17 is controversial. On one hand it can slightly improve the
>  files propagation (mostly for peers that ask chunks in fixed order, like the
>  old mldonkey behavior, or under very high upload pressure), but on the
>  other hand some slow uploader can "block" all other uploaders that want the
>  same chunk (or part of the same chunk). I can't think of a good fix that
>  would not negate the (small) benefits of the patch :(

Indeed, as you don't know exactly what the uploader already has, you
may in fact hide the only chunks it needs. Maybe another approach
could be to sort uploaders when they send the QueryChunk message, so
that the uploaders in the queue don't upload the same chunk when possible.

>  As for #25, the original remove_old_clients is broken. It uses
>  max_sources_per_file as a limit between "low on sources" and "plenty of
>  sources" behaviors, but since donkeyGlobals.ml::new_source uses a LRU
>  eviction, the number of sources *cannot* go over max_sources_per_file! Also,
>  the max_sources_age was not followed. So maybe my patch needs some
>  tuning/fixes, but something must be done for remove_old_clients in all
>  cases... (Looking at the Changelog) Oh, it was modified. I'll have to check
>  the source, then :)

I also have to check both sources to see which one is better, or if
both can work together.

>  Now for #21, I'm a bit surprized, IMO it can only improve the behavior of
>  the client (only try newly discovered sources after scheduled connections,
>  and rate-limit them), could you explain why you didn't apply the patch ?

Idem as above. I think the patches that were applied were those that
could not break the client behavior, while those which are more
fundamental have NOT YET been applied, since they have to be examined
carefully before. Probably just a question of time.

- MLDonkey



reply via email to

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