The MPLS-OPS Archive

[subodh <subodh@samsung.com>]

----- Original Message -----
From: "Ronald & Serena Travasso" <serenaronald@bigpond.com>
To: <bhavesh_modi@da-iict.org>; <mpls-ops@mplsrc.com>
Sent: Tuesday, September 30, 2003 3:13 PM
Subject: [MPLS-OPS]: IP Routing vs MPLS label switching


> Can anyone help me with a small problem.
>
> I have just started on a report  on " Conventional IP routing vs MPLS
label
> switching" with a particular reference to table lookup delay.

##MPLS concept started with fastforwarding, compare to IP forwarding but
this concept no longer exits
with the emergence of gigbit and terabit IP routers.
In most of the routers forwarding is done at the hardware (ASIC based).
Latest trend in H/W forwarding is Network processor (programable).
With H/W forarding using Network processor, difference in IP and MPLS
switching is negligible.
Now a days MPLS is mainly used for TE and VPN application.


> This is where I have a small problem which I sincerely hope you somone
would
> take the time to help me out.
>
> Here it is:
>
> I've got a Pentium IV 3GHz machine to be used as a core router maintaining
> 1000 entries in its routing table, and a visit to an entry in the routing
> table takes 1 ms.
>
> My main query is:
>
> "How would you calculate the table lookup delay for an incoming packet if
> conventional IP routing is used, and what would the table lookup delay be
if
> MPLS label switching is used."
>
> These are my thoughts:
>
> Table lookup delay depends on a number of factors - size of the routing
> table, speed of the processor where the lookup is taking place and the
> lookup algorithm. (cisco ref)
> I've been through a number of algorithms ( Patricia Tree, binary search
> etc. ) and I've noticed that the worst case scenario is O(WlogN) where N
is
> the number of prefixes and W the depth of the tree. (Gupta, Varghese) You
> have to go through the entire routing table of 1000 (W) entries to find
the
> longest match. I'm not sure if I'm on the right track here. Where does the
> speed of the processor fit in here??
>
> For MPLS when a packet with label 123 is received, the router simply has
to
> lookup line 123 of it's label forwarding table. So where's and what could
be
> the delay here?

##Label are fixed size for a given technology. In case of ethernet (shim
header)-20 bit label is used.
IP address is ---32 bit (Pfx length varies). Its a matter of 20bit search
and 32bit search (warst case).
So it depends on which algorithm is being used for search.
Same MPLS tunnel could be used for multiple destination. which could reduced
the MPLS table size.
This difference can be observed in router with CISC processor but not in H/W
forwarding.

> Further as I understand, MPLS label switching might not make routing
faster
> but definitely efficient, in terms of reduced route lookup time. If one
has
> large number of routes in the forwarding table and next hop for all of
them
> is same, I understand that one single label replaces this entire table.
>
> Any help with calculating the lookup delay would be very helpful.   It's a
> small but integral part of my report which would ensure its
completedness -
> comparison of IP routing vs MPLS label switching.
>
> Looking forward to your reply, and once more I would like to take this
> opportunity to thank you for taking the time to read this e-mail.
>
>
> Thank you and have a good day.
>
> Ronald Travasso
>
> -------
> The MPLS-OPS Mailing List
> Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
> Archive: http://www.mplsrc.com/mpls-ops_archive.shtml
>

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml