Hi, I have a Redhat Linux 7.2 box with stock 2.4.7 kernel and I'm trying to do some bandwidth management with tc.. I have created a CBQ class-hierarchy, and I have a class which is limited to 16Kbit and it can grow up to 32Kbits if there is available free bandwidth in the other classes. This class hasn't got any qdiscs attached to it, so (as I understood) it should have the pfifo_fast qdisc by default. But I have made a little test: I've mangled the TOS field of the router's outgoing ICMP packets to be 0x10 (minimize delay), and I checked with tcpdump that they are actually getting mangled. The problem is, that it doesn't seem to make any difference... the ping is fine if the class is empty, but if I'm doing a download and pinging the router at the same time, I experience delays in ping replies in both cases (either if the ICMP headers are mangled or not). Shouldn't the TOS mangled ping replies arrive with no delays at all, while the class is full with bulk traffic? Or did I misunderstood something? Thanks in advance, -- Balazs Miklos <mbalazs@n-soft.net>
On Tuesday 27 May 2003 14:09, Balazs Miklos wrote: > Hi, > > I have a Redhat Linux 7.2 box with stock 2.4.7 kernel and I'm trying to > do some bandwidth management with tc.. > I have created a CBQ class-hierarchy, and I have a class which is > limited to 16Kbit and it can grow up to 32Kbits if there is available > free bandwidth in the other classes. This class hasn't got any qdiscs > attached to it, so (as I understood) it should have the pfifo_fast qdisc > by default. But I have made a little test: I've mangled the TOS field of > the router's outgoing ICMP packets to be 0x10 (minimize delay), and I > checked with tcpdump that they are actually getting mangled. > > The problem is, that it doesn't seem to make any difference... the ping > is fine if the class is empty, but if I'm doing a download and pinging > the router at the same time, I experience delays in ping replies in both > cases (either if the ICMP headers are mangled or not). Shouldn't the TOS > mangled ping replies arrive with no delays at all, while the class is > full with bulk traffic? Why not doing the test after you add a pfifo_qdisc to that class? Just to see if it makes any difference. Stef -- stef.coene@docum.org "Using Linux as bandwidth manager" http://www.docum.org/ #lartc @ irc.oftc.net