[nsp] c6k QOS, pfc marking

From: Kinczli Zoltán (Zoltan.Kinczli@Synergon.hu)
Date: Thu Jun 20 2002 - 08:17:27 EDT


Hello,

   I'm trying to input police on a FE port on cat6500 (MSFC2, PFC2).
The FE port is an untrusted access port (i.e non-trunk). The FE port is
dedicated to a server.

  The traffic from the server is destined to a client, who is located
behind a
128k leased line.

  I'm trying to mark part of the traffic (say bellow 64kbps) with a high
precendence, say 3,
while leaving the 0 precendence for traffic above 64kbps. The leased
line is connected to
a nearby 7200, where CBWFQ will handle congestion.
 
  I'm applying a policer like this:
!
mls qos
!
ip access-l 142
 perm ip any 10.1.0.0 0.0.255.255
!
class-map test-c
 match access-g branch
!
policy-map test-pol
 class branch
 police 64000 1500 3000 128000 conf set-prec-t 3 exc tra viol drop
!
int fast4/16
 serv in test-pol
end
 
  CIR 64k, normal burst 1500B (1*MSS), excess burst 3000B (2*MSS), PIR
128k

  The RTT for a max sized TCP segment (@128kbps, again the speed of the
leased line is 128k) is 1/5 sec
so, without any policer the server/application sends 2*MSS in each 1/5
second, thus 10*MSS per second.
That equals to 10 * 12000b == 120000 kb, pretty fine matching the speed
of the leased line.

  So, i would like to have the first segment/packet be transmitted with
high precence, and have the second
segment/packet be transmitted with the deafult precendece of 0.

  But the policer seems either to set-prec to a certain value, either to
drop, depending on the normal
and maximum burst size specified. There are no packets transmitted with
precedence of zero, i.e. as
all the traffic were either conforming either violating, but not
exceeding. But, the second segment should
be exceeding.

  Anybody out there who implemented similar things before? What am i
missing here?
The c6k runs 12.1(8a)E4. I'm checking the bugnav, to see if i should
upgrade, or maybe i'm
upgrading anyway.\

Thanks in advance
  --zoltan



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:13:48 EDT