Thomas Eklund wrote:
> I also saw Steve Blakes comment that is not recommended to use TCP/UDP ports
> in the hash key for ECMP but the one of the biggest problem with the
> bundling is to send one TCP flow over different path since it might end up
> in packet re-ordering due the different delays etc and trigger TCP slow
> start. To me it makes a lot of sense then to include the TCP/UDP ports when
> you have your "hash key" since you would like that the same TCP flow take
> the same path to not experience these problems.
If you only hash on <srcaddr, dstaddr>, then all of the flows between the
two end-points will follow the same path.
There is a good paper by Cao, Wang, and Zegura analyzing the performance
of various hashing schemes at http://www.ieee-infocom.org/2000/papers/650.ps.
Regards,
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Steven L. Blake <[EMAIL PROTECTED]>
Ericsson IP Infrastructure 919-472-9913
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page: http://playground.sun.com/ipng
FTP archive: ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------
