[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Mldonkey-users] [pango] log auto_max_sources_per_file to stdout too
From: |
Pierre Etchemaite |
Subject: |
Re: [Mldonkey-users] [pango] log auto_max_sources_per_file to stdout too |
Date: |
Tue, 31 Dec 2002 22:32:45 +0100 |
Le Tue, 31 Dec 2002 18:13:12 +0100, Sven Hartge <address@hidden>
a écrit :
> I am using the 2.02 with pango-20021231a and I like the
> auto_max_sources_per_file very much.
It seems to give good results, even if it is too pessimistic about the
number of sources mldonkey can schedule.
For example, I have max_clients_per_second set to default value of 5, and if
I look at the number of sources mldonkey is scheduling (waiting for the
"Next minute: ## 2 minute: ## ... waiting for breaks: ##" messages in the
console output), I get something like 2.35 sources scheduled per second.
They're several reasons I can think of:
* the first one was that sources can be shared by several files. Of
course, all queries will be send at once, so only one connection will be
scheduled. How many sources are shared depends on the files begin
downloaded.
* not all sources are scheduled every min_reask_delay. Sources that could
not be reached, or that have no interesting chunk, will be scheduled less
often (backoff). This must happen more often if you have keeping "old
sources" in your lists, for example for rare files.
* some peers will connect to us before we connect to them. Those connections
won't need to be scheduled (even if they'll use just as much bandwidth as
if they were). I think this is the main reason why the right value for
max_sources_per_file is wrong, the algorithm tries to keep the number of
scheduled connections + the number of incoming connections below
max_clients_per_second, instead of just the number of scheduled
connections.
...