[LARTC] [OT] Shaping and squid question

[LARTC] [OT] Shaping and squid question

[Kenneth Kalmer]

List

Out of interest, I use transparent squid proxies for caching websites,
and I shape the clients that browse the net, this all works perfectly
so no questions in this regard...

CLIENT <--- eth0 ----> SQUID <----- ppp0 -----> NET

What I just realised was that all egress traffic on eth0 gets shaped,
squid gets data from ppp0 and hence is not shaped. Does this mean that
squid recieves the data at full speed and then struggles to pass the
data to the client who is shaped, thus becoming the bottle neck?

Just wondering if I should apply some ingress shaping specifically for
squid, to prevent squid from hogging all the bandwidth while
downloading a large file...

Thanks in advance

-- 

Kenneth Kalmer
kenneth.kalmer@gmail.com
http://opensourcery.blogspot.com
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] [OT] Shaping and squid question

[Andreas Klauer]

On Wednesday 04 May 2005 13:25, Kenneth Kalmer wrote:
> Does this mean that squid recieves the data at full speed and then
> struggles to pass the data to the client who is shaped, thus becoming the
> bottle neck?

If you shape everything on eth0 down to dialup speeds, including pure LAN 
traffic, then the answer is yes, you have an unnecessary bottleneck there, 
because the traffic caused by squid on ppp0 is not shaped at all. 

I'm assuming that squid traffic looks like locally generated traffic to 
iptables/QoS. I don't really use squid, so I can't check.

HTH
Andreas
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] [OT] Shaping and squid question

[Kenneth Kalmer]

On 5/4/05, Andreas Klauer <Andreas.Klauer@metamorpher.de> wrote:
> On Wednesday 04 May 2005 13:25, Kenneth Kalmer wrote:
> > Does this mean that squid recieves the data at full speed and then
> > struggles to pass the data to the client who is shaped, thus becoming the
> > bottle neck?
> 
> If you shape everything on eth0 down to dialup speeds, including pure LAN
> traffic, then the answer is yes, you have an unnecessary bottleneck there,
> because the traffic caused by squid on ppp0 is not shaped at all.
> 
> I'm assuming that squid traffic looks like locally generated traffic to
> iptables/QoS. I don't really use squid, so I can't check.

Though so. squid traffic does look like locally generated traffic for
the shaping, but I shape by marking packets so squid's return data to
the client is also properly marked and shaped together with the other
protocols not handled by the transparent squid...

Here comes my first experiences with IMQ...

Thanks for the confirmation!

> 
> HTH
> Andreas
> _______________________________________________
> LARTC mailing list
> LARTC@mailman.ds9a.nl
> http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc
> 


-- 

Kenneth Kalmer
kenneth.kalmer@gmail.com
http://opensourcery.blogspot.com
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc