The burst value depend on whether you are:
1) Implementing admission control, per the traffic profile
specification. In that case you should know the CIR/PIR/Burst values
from the contract or figure them out from application requirements
(E.g. from the number of simultaneous voice calls, as voice flows have
pretty strict profiles).
2) Implementing traffic policing for the purpose of rate-limiting.
Even though traffic-shaping should be used for that, policing is much
more resource-effective and thus widely preferred.
When using policing for rate-limiting, the burst size generally
depends on the application that you want to police. Since mostly
likely you will be using TCP-based applications, you should adjust the
burst size to accomadate for the largest burst size that could occur
in the traffic. Unfortunately, this could be only estimated in case of
single-flow traffic, based on maximum TCP windows size. Mixed traffic
requires some experimental performance estimations to be able to pick
up a better value.
You may look at the following article to get basic idea of the
measurement process:
http://www.cnaf.infn.it/~ferrari/tfngn/cat6500/police/tcp/
Or simply start with the burst of 1.5 * BitRate ... 2*BitRate as an
initial value and increment it based on the experiment results. Run
test traffic and see if the resulting rate is close to the desired
CIR. Of course, you would never be able to get to full CIR with TCP
traffic, but you could get close, at least. Keep in mind that larger
burst sizes allow for more variation in traffic profiles and thus
sudden traffic spikes. This, in turn, may result in temporaty
congestions in the network core. Therefore, pickuing up the optimal
burst size is always a balance between an optimal CIR and stable
network.
-- Petr Lapukhov, petr_at_INE.com CCIE #16379 (R&S/Security/SP/Voice) Internetwork Expert, Inc. http://www.INE.com Toll Free: 877-224-8987 Outside US: 775-826-4344 2010/4/15 Persio Pucci <persio_at_gmail.com>: > hi folks, > > I am trying to determine burst sizes for policies to be used on a Cat6500 > using mls qos aggregate-policer, but I am not confident on the numbers I am > getting. > > For instance, for a 1mbps (1024000bps) policy, I would use a CIR of 1024000 > bps. The Cat6500 literature ( > http://www.cisco.com/en/US/docs/switches/lan/catalyst6500/ios/12.2SXF/native/configuration/guide/qos.html#wp1726251) > says > that the Tc is 1/4000sec (0.25ms). For that Tc, I would have to use: > > CIR=Bc/Tc > > 1024000=Bc/(1/4000) > > Bc=1024000*(1/4000) > > Bc=256 > > > That being 256 bits. Considering the normal burst size field asks for a > value in bytes, that makes 256/8=32 bytes, making my config line: > > mls qos aggregate-policer 1mbps 1024000 32 32 conform-action transmit >> exceed-action drop violate-action drop > > > The burst value does not sound reasonable for me. Wanted to check with you > guys if I am missing something here. > > Persio > > > Blogs and organic groups at http://www.ccie.net > > _______________________________________________________________________ > Subscription information may be found at: > http://www.groupstudy.com/list/CCIELab.html Blogs and organic groups at http://www.ccie.netReceived on Mon Apr 19 2010 - 12:58:36 ART
This archive was generated by hypermail 2.2.0 : Sat May 01 2010 - 09:49:57 ART