Hi all, Just finished reading and re-reading the advanced routing how-to. This this is marvelous!!! Thank you all that spent your time and devoted it to such a great paper! However there is something that I am missing from the picture. By default each interface is bundled with the pfifo_fast that prioritizes the packets based on TOS. If for say we do this: # tc qdisc add dev eth0 root tbf rate 100kbit latency 100ms urst 2048 Do we actually lose the whole point of TOS? Or there is stil an underlying filter somewhere in the kernel that will take care of the TOS scheduling just like pfifo_fast did? Also in the following example: # tc qdisc add dev eth0 root handle 1: prio We automatically create three handles 1:1 1:2 1:3, with the default priomap. However what happens to traffic that goes down 1:1 if there is no qdisc attached to it? Does it get dequeued right off the back or there is again something similar to pfifo_fast waiting for it? Cheers! _______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
On Monday 12 May 2003 15:35, Peter Rabbitson wrote:> Hi all, > Just finished reading and re-reading the advanced routing how-to. This this > is marvelous!!! Thank you all that spent your time and devoted it to such a > great paper! However there is something that I am missing from the picture. > By default each interface is bundled with the pfifo_fast that prioritizes > the packets based on TOS. If for say we do this: # tc qdisc add dev eth0 > root tbf rate 100kbit latency 100ms urst 2048 Do we actually lose the whole > point of TOS? Or there is stil an underlying filter somewhere in the kernel > that will take care of the TOS scheduling just like pfifo_fast did?You loose the TOS settings.> Also in the following example: > # tc qdisc add dev eth0 root handle 1: prio > We automatically create three handles 1:1 1:2 1:3, with the default > priomap. However what happens to traffic that goes down 1:1 if there is no > qdisc attached to it? Does it get dequeued right off the back or there is > again something similar to pfifo_fast waiting for it?There is always a qdisc responsible for sending the packets. If you add a prio qdisc, you also created 3 small fifo qdiscs for each class. If you want some example shaping scripts (mostly cbq and htb), you can go to docum.org. Stef -- stef.coene@docum.org "Using Linux as bandwidth manager" http://www.docum.org/ #lartc @ irc.oftc.net _______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
So in other words if we define a cbq at 1:0 and then define a class 1:1, and DO NOT attach any qdiscs at parent 1:1, all packets that went to 1:1 will be simply discarded? P.S. Examples are fine but going down and getting dirty is way more fun :) On Mon, May 12, 2003 at 09:18:48PM +0200, Stef Coene wrote:> On Monday 12 May 2003 15:35, Peter Rabbitson wrote: > > Hi all, > > Just finished reading and re-reading the advanced routing how-to. This this > > is marvelous!!! Thank you all that spent your time and devoted it to such a > > great paper! However there is something that I am missing from the picture. > > By default each interface is bundled with the pfifo_fast that prioritizes > > the packets based on TOS. If for say we do this: # tc qdisc add dev eth0 > > root tbf rate 100kbit latency 100ms urst 2048 Do we actually lose the whole > > point of TOS? Or there is stil an underlying filter somewhere in the kernel > > that will take care of the TOS scheduling just like pfifo_fast did? > You loose the TOS settings. > > > Also in the following example: > > # tc qdisc add dev eth0 root handle 1: prio > > We automatically create three handles 1:1 1:2 1:3, with the default > > priomap. However what happens to traffic that goes down 1:1 if there is no > > qdisc attached to it? Does it get dequeued right off the back or there is > > again something similar to pfifo_fast waiting for it? > There is always a qdisc responsible for sending the packets. If you add a > prio qdisc, you also created 3 small fifo qdiscs for each class. > > If you want some example shaping scripts (mostly cbq and htb), you can go to > docum.org. > > Stef > > -- > > stef.coene@docum.org > "Using Linux as bandwidth manager" > http://www.docum.org/ > #lartc @ irc.oftc.net > > > _______________________________________________ > LARTC mailing list / LARTC@mailman.ds9a.nl > http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/_______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
On Monday 12 May 2003 23:08, Peter Rabbitson wrote:> So in other words if we define a cbq at 1:0 and then define a class 1:1, > and DO NOT attach any qdiscs at parent 1:1, all packets that went to 1:1 > will be simply discarded?No. By default a fifo qdisc is attached to the class. Stef -- stef.coene@docum.org "Using Linux as bandwidth manager" http://www.docum.org/ #lartc @ irc.oftc.net _______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/