Re: [Pan-devel] Inefficiencies in Pan threading?

[Jason Leach]

Pan is using all 4 connections to your server, but it is actually using
all 4 to grab each _part_ of a single file at a time. Files posted on
usenet are typically made of anywhere from 10-1000 individual text
message posts. When pan builds its article list for a group, it
automatically joins them so it looks like just one line in the article
list, but at the end of each line you'll see something like (01/24),
indicating there are 24 parts total. So when you tell pan to download
that file, it really needs to get 24 messages. Pan downloads a message
on each server connection, then when all the parts are done, combines
and decodes that one file.

The times where it looks like it's downloading multiple files are when
there are fewer than 4 parts left to get on one file (i.e. currently
downloading 22/24, 23/24, 24/24 of file #1), so Pan intelligently moves
on to downloading the first part of the next file (i.e. 01/24 of file
#2).

Some (many?) people have limited speeds per connection, say 30 k/s, but
their server allows them to connect more than once, say 4 max
connections. So, this way pan can effectively grab a single file at 120
k/s, instead of grabbing file #1 at 30k/s, file #2 at 30k/s, etc. So
single files will download faster using this connection per part
compared to doing a connection per file.

Hope that helps,
Jason.