pan-users
[Top][All Lists]
Advanced

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

Re: [Pan-users] Connection limit locked at 4?


From: Duncan
Subject: Re: [Pan-users] Connection limit locked at 4?
Date: Tue, 5 Feb 2013 01:37:26 +0000 (UTC)
User-agent: Pan/0.140 (Chocolate Salty Balls; GIT 5133e23 /usr/src/portage/src/egit-src/pan2)

Chris Gentle posted on Mon, 04 Feb 2013 15:37:03 -0600 as excerpted:

> I've noticed that Pan seems to want to limit my server connections to 4,
> although my provider allows 30 simultaneous connections.  The button in
> the GUI to increase this limit is grayed out.  I can edit the
> servers.xml file and manually adjust the number and sometimes get it to
> work but sometimes it seems stuck at 4.  Is there a reason for this?

As Heinrich says, GNKSA (good netkeeping seal of approval).  Charles 
Kerr, pan's lead/primary developer for many years, worked hard for pan's 
100% rating.

These days news works rather different, and servers generally must 
control the number of connections they allow, so the limit of four per 
server from GNKSA does seem a bit anachronistic, even for people who 
generally support the other GNKSA points.  But that'd break the 100% and 
many (including me) are afraid doing so might start a slide on strictness 
on the other points which we do strongly support, so...

The compromise has been to have pan's GUI only do the max-four check when 
the server setting change -- it always attempts as many connections as 
are set in servers.xml, without that max-four check, thus allowing 
"advanced" users (who presumably know what they are doing) to manually 
edit the config file to whatever number of connections they want to try, 
letting the server say no if that account isn't allowed that many 
connections.  That complies with GNKSA, while allowing those who want 
more connections to be able to set them.

But if you manually edit servers.xml, pan should take that number of 
connections and run with it.  The only time it might reset to four is if 
you change some other server setting from the GUI, so pan rewrites 
servers.xml, in which case it'll probably rewrite four per server and 
you'll have to manually edit servers.xml again.


Meanwhile, I'd suggest that in most cases you'll likely get best thruput 
at perhaps 10 connections, depending on the speed of the storage you're 
saving to as well as on the speed of your network connection and/or 
number and speed of CPU cores.    IOW, you'll almost certainly hit a 
bottleneck elsewhere, before you hit that 30 connections limit, and above 
that, you're just doing more work at a lower speed per connection to 
maintain the overall same or lower download speed.  Even four connections 
will likely come pretty close to hitting the bottleneck, rather closer 
than you might think.  Unless of course the server is seriously limiting 
the per-connection speeds as well, but there's little point if they're 
allowing 30 connections, as it's actually less work at the server (as 
well as the client) to run fewer connections at higher speeds, so few 
providers limit the per connection speeds much any more, and most people 
run into their general internet connection speed caps (or local disk 
write speed limitations) before they get anywhere near the 30+ 
connections commonly allowed on paid NSP accounts, these days.

Unless of course your inet connection is gigabit or better, you're 
running an 8-core or better system (with a minimum gig per core RAM), and 
saving to a fast-write SSD or many-spindle RAID-0/10...  THEN you might 
actually get increased overall speeds upto 20-30 connections...

Now that I have the terabyte block from astraweb to play with and a 6-
core anyway, with 16 gigs RAM, tho I'm still running "spinning rust" and 
cap out at ~ 20 Mbit inet, I should probably run some experiments and get 
a feel for where the bottleneck actually is these days...

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman




reply via email to

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