[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: DiffServ Testing on local LAN





> Hello All,
> 
> Can we test the performance of the Linux diffserv patch [ iproute2 and
> tc ] on a local LAN ?

Yes you can.

> ( by this I mean that can we show differences in the end to end delay
> experienced by various classes
> of traffic EF, AF, BE etc and also strict traffic isolation for premium
> traffic flows )
> 
> We have been trying this for the past many days....but though we have
> been able to achieve
> some degree of traffic isolation---using the bounded and isolated
> keywords of the "tc" tool
> we have not observed difference in the end to end delay experienced by
> the various
> traffic classes.

What kind of end-to-end measurement tool did you use? If you had used
ping, then the ICMP packets will not be marked with the correct DSCP and
so will not traverse the path through the queues that the different EF, AF
marked packets experience. 

>We use 2 machines connected to the local LAN ( 10 Mbit
> Ethernet ) with IP addresses
> 192.9.201.25(sender )  and 192.9.201.26(reciever) .Only the sender is
> diffserv enabled..so I
> presume that it should throw out / transmit EF and AF packets faster  in
> preference to BE packets.

Did you get the correct throughputs that you alloted to the streams?

Also look at the the tc -s qdisc results to see the packet drops per VQ,
this should give you a good idea of the performance.

> But we don't get the expected results......
> 
> Is there something wrong with my setup or are we making some other
> "FUNDAMENTAL" error.
> We have tried many different configurations of qdiscs, classes etc using
> tc....but have not got
> anything near the expected results......(i.e the behaviour of the
> various classes as defined in the RFCs and
> the paper "Differentiated Services on Linux" by Almesberger, Jamal et
> al)
> 
> Please advise...I seem to have reached a dead -end !!!!

Try using a tool like say Tcpdump. It plots the throughput, Round trip
time, Congestion window size, segment size, etc on a per tcp connection
basis. It uses a tcpdump file as an input. It will give you a good idea of
the end-to-end delay and tcp dynamics experienced by the different DSCP
marked packets. 

It is one of the several tools recommended in the Informational RFC 2398
"Some Testing Tools for TCP Implementors"

We used it in our paper matching AF to VBR available from

http://www-networks.tip.csiro.au/~mminhazu/dif_serv.html

Cheers
Muneyb.
_________________________________________________________
Muneyb Minhazuddin - Telecommunications Research Engineer    
CSIRO Telecommunication and Industrial Physics                
Marsfield, NSW, Australia.                                    
                                                             
Phone no. : 61 2 9372 4113
FAX       : 61 2 9372 4490          
e-mail    : mminhazu@tip.csiro.au  
Home Page : http://www-networks.tip.csiro.au/~mminhazu
---------------------------------------------------------