LARTC - Feb 2002 - TC with simple PRIO and TBF setup. Please help.

I have a simple dream. That dream is to prioritize traffic on my outbound
cable modem link so that I can run a 24/7 webserver AND a 24/7
counter-strike server. The game server is seldom full, so I want unused
bandwidth to go to the web server, while the web server takes a huge dive
in speed when the game server fills up. Finally, a small slice must be
available to the two user machines with web browsers... assuming
they''re
not playing counter-strike.

 I created a simple three band prio/tbf setup to implement this, but the
game server still gets lagged to hell when it is moderately full and
someone starts a download from the web server. I''ve place the webserver
in
band 2, lowest priority, which *I thought* would prevent this from ever
happenning.

 Will someone more gifted than I please look at the following script for
me and point out the error. The upstream on the cable modem is supposed to
be 384kbit, I''m using 320kbit to be conservative. Please tell me what I
need to do to get this to work.

   -=Chris (I can''t tell you how happy this would make me)

### MY TC script
#!/bin/sh

#Add the root PRIO type qdisk, three priotiy bands, default band 1
tc qdisc add dev eth0 root handle 1: prio \
        bands 3 priomap 1 1 1 1 1 1 1 1 1 1 1 1 1 1

#1:1 is band 0 (highest priority). Give it most of the bandwidth if
#needed and tollerable game latency
tc qdisc add dev eth0 parent 1:1 handle 10: tbf rate 280kbit buffer 2500\
        latency 300ms

#1:2 is band 1 (mid priority, default band). This is mostly browser
#traffic, give it very little of the upstream pipe and good turnaround
tc qdisc add dev eth0 parent 1:2 handle 20: tbf rate 45kbit buffer 2500\
        latency 100ms

#1:3 is band 2 (lowest priority). This level should be given the entire
#pipe if the other two bands don''t need it. This is large tranfers,
high
#latency is fine
tc qdisc add dev eth0 parent 1:3 handle 30: tbf rate 320kbit buffer 2500 \
        latency 600ms

#Large transfers come from the web server, send it to band 2
tc filter add dev eth0 parent 1:0 protocol ip u32 match \
        ip sport 80 0xffff flowid 1:3

#The game server runs on udp port 27015, send it to band 0
tc filter add dev eth0 parent 1:0 protocol ip u32 match \
        ip sport 27015 0xffff flowid 1:1

On a quick look at your configuration, it seems that you''re not slowing
down the packets as they leave eth0 and enter your cable modem. You''re
doing this on each prio qdisc with a tbf, but you need a root class based
qdisc so that you can slow down traffic at the root level and have the
packets queue properly. Something like this:

-Root- CBQ @ 320Kbit bounded
	-3 Band Prio-
		-SFQ-
		-SFQ-
		-SFQ-

This will make your latency go *way* down provided the value of 320Kbits
works in your situation. Experiment to find a good value. Also, you may
want to consider adding an ingress policer to your setup so that latency
doesn''t jump way up when someone on your internal network starts a
large
download. For tips on getting this all to work, see the advanced routing
howto section on the "holy grail" of a QoS setup, or email me directly
(off the list) and I''ll be glad to help you out further.

-Ross Skaliotis

On Tue, 19 Feb 2002, Chris wrote:
>
>  I have a simple dream. That dream is to prioritize traffic on my outbound
> cable modem link so that I can run a 24/7 webserver AND a 24/7
> counter-strike server. The game server is seldom full, so I want unused
> bandwidth to go to the web server, while the web server takes a huge dive
> in speed when the game server fills up. Finally, a small slice must be
> available to the two user machines with web browsers... assuming
they''re
> not playing counter-strike.
>
>  I created a simple three band prio/tbf setup to implement this, but the
> game server still gets lagged to hell when it is moderately full and
> someone starts a download from the web server. I''ve place the
webserver in
> band 2, lowest priority, which *I thought* would prevent this from ever
> happenning.
>
>  Will someone more gifted than I please look at the following script for
> me and point out the error. The upstream on the cable modem is supposed to
> be 384kbit, I''m using 320kbit to be conservative. Please tell me
what I
> need to do to get this to work.
>
>    -=Chris (I can''t tell you how happy this would make me)
>
> ### MY TC script
> #!/bin/sh
>
> #Add the root PRIO type qdisk, three priotiy bands, default band 1
> tc qdisc add dev eth0 root handle 1: prio \
>         bands 3 priomap 1 1 1 1 1 1 1 1 1 1 1 1 1 1
>
> #1:1 is band 0 (highest priority). Give it most of the bandwidth if
> #needed and tollerable game latency
> tc qdisc add dev eth0 parent 1:1 handle 10: tbf rate 280kbit buffer 2500\
>         latency 300ms
>
> #1:2 is band 1 (mid priority, default band). This is mostly browser
> #traffic, give it very little of the upstream pipe and good turnaround
> tc qdisc add dev eth0 parent 1:2 handle 20: tbf rate 45kbit buffer 2500\
>         latency 100ms
>
> #1:3 is band 2 (lowest priority). This level should be given the entire
> #pipe if the other two bands don''t need it. This is large
tranfers, high
> #latency is fine
> tc qdisc add dev eth0 parent 1:3 handle 30: tbf rate 320kbit buffer 2500 \
>         latency 600ms
>
> #Large transfers come from the web server, send it to band 2
> tc filter add dev eth0 parent 1:0 protocol ip u32 match \
>         ip sport 80 0xffff flowid 1:3
>
> #The game server runs on udp port 27015, send it to band 0
> tc filter add dev eth0 parent 1:0 protocol ip u32 match \
>         ip sport 27015 0xffff flowid 1:1
>
>
> _______________________________________________
> LARTC mailing list / LARTC@mailman.ds9a.nl
> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/lartc/
>