The MPLS WG Archive

["Brijesh Kumar" <brijesh.kumar@att.net>]

Nitin:

You need to analyse your packet forwarding code and the system configuration
with regards to:

1. What kind of L3 route look up is implemented? Is there any significant
difference in the routing table size at the ingress and the egress node? In
any case, measure time for route look up at ingress.

2.  Make sure you add a specific route to destination instead of forwarding
using a default route. This is to make sure you get average rather than
worst case performance from whatever loop up algorithm you are running.

3. Also, have you set up any IP packet classification or access control
policies at ingress? Check that the code does not do some default packet
classification or apply any default access control policies when you have
configured none? Both of these activities are very compute intensive and can
substantially impact the performance. If that is the case, add suitable
instrumentation  code to your packet forwarding code.

If you do the above, you are more likely to get the cause of long processing
time at Ingress. It  looks to me that you are unlikely to get similar
results if you were to use some other implementations of the forwarding
code. Try some other systems/implementation and see how it compares with
your current results.

Cheers,

--brijesh


----- Original Message -----
From: "nitin p" <nitin_panj@yahoo.com>
To: <GAlmaalouf@aol.com>; <eosborne@cisco.com>; <jshen@cad.zju.edu.cn>
Cc: <rameshbabu_j@infosys.com>; <mpls-ops@mplsrc.com>; <mpls@UU.NET>
Sent: Monday, December 09, 2002 6:24 AM
Subject: Re: [MPLS-OPS]: Jitter and MPLS


> Hi George,
> I have posted my results in of the mails, but here
> they are again:
>
> Ingress: 3.665 milli seconds
> Egress: 0.035 milliseconds
> core:    0.018 milliseconds
>
> IP: 0.023 milliseconds.
>
> For my measurements I have attached a sniffer at both
> incomming and outgoing interface (same sniffer for
> both so no clocking issue). I did the same thing for
> boths LSR and LERs. Question of number of LSRs should
> not affect the results, considering my measuring
> technique.
> Jing said He is also getting similar results, I am not
> sure how he has done his measurement.
>
> Thanks,
> Nitin
>
>
>
> --- GAlmaalouf@aol.com wrote:
> > Nitin-
> >
> > It would be much helpful if you can specify the
> > delay in a quantitative
> > manner. This way we can really understand what are
> > you talking about in term
> > of delays.  Also, the specific number of Hops or LPS
> > router, how many egress
> > and ingress steps involved in the calculation. A
> > general observation would
> > not  be very accurate, since the implementation
> > would be very different from
> > one application versus the other, and from one type
> > of services versus
> > another.
> >
> > Thanks-
> >
> > George
> >
> >
> >
>
>
> __________________________________________________
> Do you Yahoo!?
> Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
> http://mailplus.yahoo.com
>
>