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

Confusion regarding TC (cbq), policing and bursts



Hi,

Problem understanding what exactly is going on here.  I have a series of
ingress traffic streams (9 of them, in one group), each destined for a
remote system.  On my egress NICs, I've created cbq qdiscs, classes and
appropriate filters to pick the packets.  This works.  The problem I have is
that the streams are set such that each will deliver a quantity of data (say
200kbit, which comes to about 390.6 64-byte packets).  If the stream group
is to complete the sending of the data within one second, the effective rate
would be 1800kb/s (200kb/s x 9).  So, I essentially have 200kbit of data
coming through at 1800kb/s in one burst.  Each stream acts in this manner.  

Initially, I set 'tc class .. cbq .. rate 200kbit .. maxburst 20', but this
doesn't seem right.  We actually see more data coming though since the
traffic is so bursty.  

Questions:

1.  I've gleaned from the net, that 'maxburst' refers to the number of bytes
allowed in one burst.  Setting this value to much more than 100 seems to
result in high packet loss.  But only 100bytes per burst?  Seems awfully
small.

2.  I've modified as per good advice found here and elsewhere, the PSCHED
(now set to CPU) and HZ (now 1024) values.  What then, do those values mean?
Why is one value better than another?

3.  What is the sampling rate, ie:  how often does tc look at the packet
stream to determine if a drop/accept action should occur?

Cheers!
Jon

---
Jonathan Earle
Linux Admin, ONC Software Development
Nortel Networks
* Phone: 613-768-1371  ESN: 398
* Fax:   613-768-4195  ESN: 398
* Cell:  613-612-0946
* Email: jearle@nortelnetworks.com