[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
It all works as expected. I generate traffic of 6 Mbps marked 0x88, the
packets are remarked to 0x88, 0x90 and 0x98. The policing works.
How it worked? Its a bit bizarre
The weird thing was, I was using the program "ttt" on the ingress
interface of the edge router and facing all the problems that I have been
reporting. It froze this morning, so I killed it, Lo Behold! my packet
remarking started to appear on my tcpdump output.
I used ttt on the egress interface, however that didn't affect the
policing. So my guess is the ttt implementation is to blame and somehow
interferes with the ingress implementation.
I have used ttt before with other scripts (without ingress) on the same
interface without any problems.
My throughput results now match yours.
On Tue, 15 Feb 2000, jamal wrote:
> I have tested this particular script and it works. This was a while back
> though. Luckily i took notes and they just coincidentally happen to be
> right next to me ;->
> On Wed, 16 Feb 2000, Muneyb Minhazuddin wrote:
> > I have checked that the ingress and egress interfaces are correctly
> > set. The problem still exist.
> > _________ ______________ ____________
> > | |eth0 eth0| Edge Router |eth1 | |
> > | Source |----------->| Edge31-ca-u32|------>|Destination |
> > |_________| |______________| |____________|
> > INDEV = eth0
> > EGDEV = eth1
> > Source single ttcp source from source to Destination. I have added an
> > option -q to ttcp that sets the tos byte to the value given on the command
> > line.
> > Monitored output with ttt and verified the tos byte value by running
> > tcpdump on Edge router's eth1(Egress) interface and the destination.
> You are correct about getting close to 6Mbps; My notes show around
> 5.1Mbps. (this is the sum of the meters in the ingress). To see a drop,
> fix the BE for AF41 so that it drops the packets instead of demoting them
> to BE. With the fixed script, i get around 4.2Mbps
> The difference in our metering results is most probably related
> to the clock sources; mine is based on the pentium TSC -- by default
> you use the system clock which runs at a granularity of 10ms. The
> clock is not such a big deal though if you have a constant packet arrival.
> So there could be other factors;
> i used tcpblast with 1000 byte packets, you used tttcp etc
> > They
> > are still marked to 0x88 (all of them). The problem still exist.
> Check carefully. Most of them are marked as 0x88, but not all of them.
> Some more tests i ran:
> -tos 0x90 at source: throughput around 3.6 Mbps (fixed script 2Mbps)
> -tos 0x98: 2.5Mbps (fixed script 1Mbps)
> -tos 0x0 for BE or any other TOS not being filtered: 1,5 Mbps (1.6Mbps)
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 : email@example.com
Home Page : http://www.atnf.csiro.au/~mminhazu