[c-nsp] BGP Balanced

[Rodney Dunn]

On Wed, Aug 25, 2004 at 10:42:46AM -0700, Bruce Robertson wrote:
> Thank you for that excellent explanation, Rodney!
> 
> One question, though... I've always used per-packet when I have, for example,
> four T1s, because it's been my understanding that with the src/dst hash,
> an individual user will be limited to the bandwidth of a single T1 during
> a particular transfer, because the packets will always follow the same path.
> With per-packet, the transfer can use more of the available bandwidth.  I have
> observed this behavior in practice, so the out-of-order packets can't be
> causing *too* much trouble.  Could you comment, please?

You are correct that a single user doing a transfer to a single
destination will always take the same path.

I'll guarantee you that on most applications there is *some* performance
hit. 

Just try it.  Take two routers in the lab and put them back to back
with 4 serial cables with a single FTP going between two clients.

Compare the transfer rates with the different options.

The best you'll get is with MLPPP because it solves the 
out of order problem and the "only use one link" problem.

However, that single transfer will require the ability for
the transfer to take the full BW of a single link and that
may not be possible depending on the other links in the path.
It could be though.  It just depends on the full setup.

I read a report one time from IEE about packet
reordering affect on the Internet.

Sept/Oct issue of IEEE Magazine named:

The Effect of Packet Reordering in a Backbone Link on Application
Throughput 


Rodney


> 
> --
> Bruce Robertson, President/CEO				     +1-775-348-7299
> Great Basin Internet Services, Inc.			fax: +1-775-348-9412
> http://www.greatbasin.net
>