OSEC

Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com
 
Re: Internet Draft : Prioritized Treatment of Specific OSPF Packets and Congestion Avoidance #1

From: Acee Lindem (aceeREDBACK.COM)
Date: Wed Apr 02 2003 - 18:08:04 CST

Mitchell,

This draft has been in progress for some time. There is some
concensus that many of the techniques are the right things to
do to improve scalability (though possibly not all of them).
Much of what you're proposing below is either new or dictates
a specific implementation of the techniques in the draft.
I don't think we want to add any of this to the BCP at
this time.

Thanks,
Acee

Erblichs wrote:
> Group,
>
> Lets keep it simple..
>
> A) 2) The Proposal
>
> (1b) Separate the hello packets into its own FIFO queue
> for later processing. Periodicly process hellos
> based on the hello interval and the number of
> nbrs on the interface.
>
> Upon recieving the event that informs that a adj may
> be down due to hellos, mark the number of hellos
> currently in the queue.
>
> Process all the hellos up to the mark and identify
> whether the event is cleared. If not, do normal
> processing that takes down the nbr, adj.
>
> This way hellos can be processed as "lower priority".
>
> (1c) Packet Pacing
> Due to the fact that packet congestion on a interface is
> based on packet spacing with interpacket gaps (ala Ethernet),
> dynamicly adjust the rate of outgoing packets based on
> multicast or unicast destination, the number of nbrs/adjs
> on the interface, and the bandwidth of the interface.
>
> The algorithm that determines spacing should have a defined
> rampup speed, and have a exponential slowdown if congestion
> is determined (tcp Van Jacobson's algorithm is a good start)
>
> (1c) Router capability and Congestion determination should
> be based on past round-trip
> times of known events with the nbr; ie, Initial Database
> Synchronization packets.
>
>
> These are just 3 of dozens of items that can be used to
> determine and adjust for increased load/scalability.
>
> Done.. :)
>
>
> Mitchell Erblich
> Sr Software Engineer
>

--
Acee