In article <bhvvs3$2g0b$(E-Mail Removed)>, Martin Greco wrote:
> Hi everyone
>
> I have a RedHat 8 Box as gateway for about 15 clients through NAT. The
> connection is a 256Kb DSL over pptp.
> the situation is this: when eth1 (which is the internal NIC) is down I have
> about 26 KB downstream and 12 KB upstream in several tests, but when it is
> up and there are many clients operating the performance goes down, that's it
> 5 or 10 KB downstram and 3KB or 4KB upstream.
>
> Could this be possible? is it a configuration issue or a problem with the
> connection?
>
Very possible. This is due to the nature of all internet links where TCP
traffic particularly is involved. With TCP, the client machines send
ACKnowledgement packets to tell the server it received the data fine and in
order. If the bandwidth is saturated in the opposite direction then these
ACK packets get lost due to conjestion.
The blind way to solve this is more bandwidth, the clever way Linux folk fix
it is with Quality of Service (QoS for short). What this does is 'shape' the
traffic and prioritizes the packets so that the packets which need low
latency (e.g. SSH) leave as quickly as possible whilst 'bulk' traffic (e.g.
FTP and file sharing software) is used as filler.
A good place to learn are read up on this is
http://lartc.org/ - Advanced-Routing-HOWTO
<cheapo_plug>
you can also look at my website and my script as a solution to the problem in
good GPL fashion. Its a script that can be dropped in and should not effect
anything thats already in place on the firewall.
http://digriz.org.uk/jdg-qos-script/
</cheapo_plug>
Regards
Alex
Similar Threads
Linear Mode
