On Sun, Jan 28, 2007 at 03:57:48PM +0100, Ralph Passgang wrote:
> I think it could be the conntrack table.

I think that, too.

> On my asus WL500gP conntrack gets loaded with this output:
> -> ip_conntrack version 2.1 (5953 buckets, 5953 max) - 360 bytes per conntrack
> 
> This means that it can handle 5953 connections, and after this the table is 
> full and new connections will be dropped. This "max" value can be manipulated 
> via /proc/sys/net/ipv4/ip_conntrack_max, but be aware that you need some free 
> memory if you increase the amount of connections that conntrack can handle. 
> You might try to double the amount of connections, but please don't try to 
> get a million connections working on a embedded device :)

I'd (also) recommend setting the timeout lower, because the default is quite 
high (5 days). Look at /proc/sys/net/ipv4/netfilter/*, especially 
ip_conntrack_tcp_timeout_established. I guess 3600 (one hour) solves this 
problem even if you do not increase the ip_conntrack_max value.

Unfortunatelly, P2P tends to leave a lot of hanging (not properly closed) 
connections over, and having them remembered for 5 days might not be a good 
idea. Maybe we should even make this an menuconfig/wib option, otherwise a lot 
of people will encounter stalled routers while using P2P (which is quite common 
theese days). A too low value might violate TCP specs, though ... i don't know 
them.

Bye,
Lothar
_______________________________________________
freewrt-users mailing list
[email protected]
https://www.freewrt.org/lists/listinfo/freewrt-users

Reply via email to