The new IPMP docs[1] say on page 97:

   However, for a given TCP connection, IPMP uses only a single
   underlying interface to send and receive traffic.  No such
   limitation exists for link aggregation. In general, you deploy link
   aggregation to obtain better network performance, while you use
   IPMP to ensure high availability.

1. http://www.opensolaris.org/os/project/clearview/docs/ipmpov.pdf

I don't believe this is correct, based on multiple factors.  Key among
them are my experience with escalations when Sun Trunking was used
with a round-robin hashing mechanism (huge reassembly issues) and the
man page for dladm(1M).  On SXCE build 95, dladm(1M) says:

         -P policy, --policy=policy

             Specifies the port selection policy to use for  load
             spreading  of outbound traffic. The policy specifies
             which dev object is used to send packets.  A  policy
             is a list of one or more layers specifiers separated
             by commas. A layer specifier is one of  the  follow-
             ing:

This and the options that follow seem to make it fairly clear that
every packet in a given TCP stream will hash to the same outbound
port.  The key benefit of link aggregation is that you can have
inbould load spreading (but still not in the same TCP stream).

Do I misunderstand something?

-- 
Mike Gerdts
http://mgerdts.blogspot.com/
_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to