[j-nsp] CoS issue

[Amjad Ul Hasnain Qasmi]

I have some issues with configuring CoS. Below the the test bed arrangement.

(CE1)------(PE1)===MLPPP===(PE2)-------(CE2)

I m implementing CoS b/w PEs, MLPPP is configured to bundle 2 E1s. Under CoS
configuration I configured different schedulers for different class of
services like EF, AF BE. I configd 75% of transmit rate on PE egress intf to
EF services, buffer-size is configd as 1 percent, drop profile associated
with scheduler is also less aggressive.

With this configuration I can ping with a packet size of 1699B with 0%
packet loss, at this time average link utilization remains no more than
450K. As soon as I increase the packet size to 1700B, packets start
dropping. 

With same transmit rate I increases the buffer-size to 2% and observe that
now I can ping with a packet size of 1955B, although packet size of 1956B
starts dropping again.

As per my understanding, packet goes into the buffer when there is a
congestion in the link but in my case link is not congested b/c transmit B/W
is configd as 75% (i-e 2 * E1 * 0.75 = 2976Kb). There should not be any
effect of buffer-size with a packet size of 1700 or 1956B because average
utilization remains below 1000Kb, EF queue is also configd as strict high.

 

Can anyone suggest when packet comes into the buffer? What is the impact of
buffer-size on throughput, and why in my case packets are buffered while the
queue B/W is less than 50% utilized.

 

Only relevant config is shown.

 

Class-of-Services {

 

EF {

interpolate {

fill-level [ 0 90 95 100 ];

drop-probability [ 0 1 3 5 ];}

schedulers {

 

EF {

transmit-rate percent 75;

buffer-size percent 1;

priority strict-high;

drop-profile-map loss-priority high protocol any drop-profile EF;

drop-profile-map loss-priority low protocol any drop-profile EF;

}