Re: Tor client performance (was Re: URGENT: patch needed ASAP for authority bug)

[Lucky Green <shamrock@xxxxxxxxxxxxxx>]

Here are some data points that may or many not be useful to the Tor 
developers.

I run two middleman relays, Tonga and Lifuka. The two relays are nearly
identical VMs on the same ESX host. The only difference between the two 
is that Tonga is a Bridge Authority and runs Mixmaster as well.

Tonga has historically been pushing between 30 and 40 Mbps. Lifuka tends 
to push between 40 and 50 Mbps, all according to munin. I don't know why 
Lifuka tends to handle more traffic than Tonga. It has been like this 
for a ages.

Again according to munin, Tonga has about 5k active connections. That 
number has remained relatively steady over the last year. If anything, 
there has been a slight decrease between November 2009 and February 
2010. Though connections has since crept back up to the long term 
average. You can call the number of connections roughly the same for the 
last 12 months.

Lifuka on the other hand saw an average of about 8k active connections 
for much of 2009. The number of connections have been slowly creeping up 
from 8k connections in November 2009 to about 16k active connections 
today. But I see nothing in the graph that shows a recent jump.

Lifuka's connection count is more uneven than Tonga's, but the 
regression since November 2009 looks roughly linear to me. Let me know 
if you want the munin graphs.

This leaves open the question why Lifuka has double the number of open 
connections than Tonga with only about 25% more traffic, but this is not 
a new question.

Hope this helps,
--Lucky