Miguel Delapaz wrote on 08/02/2006 01:00:40 PM:

>
> Dennis,
>
> When more than one path to a destination exists, and they have the
> same "cost" (as defined by the routing protocol), z/VM TCP/IP will
> round robin packets to that destination through the available paths.
> For static routing, this behavior is controlled by the
> EQUALCOSTMULTIPATH parameter on the ASSORTEDPARMS statement.  For
> dynamic routing (MPRoute), this behavior is always enabled.  z/VM
> TCP/IP only does this on a per-packet basis (we are looking other
> options for the future).  A "hack" you can use to get around this,
> in your case, is to modify the COST0 parameter on one your
> OSPF_INTERFACE statements.  By forcing the links to have different
> costs, MPRoute will only use the path with the lowest total cost,
> and will no longer round-robin the packets.  In the event of adapter
> failure, MPRoute would fail-over to using the higher cost path.
>
> Regards,
> Miguel Delapaz
> z/VM TCP/IP Development
>
>

Miguel,

My network guy doesn't like that option very much, although we understand
thats the way TCPIP currently works.

He asks:  "Most network equipment supports either selecting one of the
paths from an ECMP environment based on an address hash or round-robin.
Apparently IBM doesn't support anything other than round-robin???  Are
there any plans to enhance this in the future?"

Are there any enhancement plans in this area?

Would a direct VSWITCH connection to the zLinux server (running RHEL AS 4)
be more flexible in this area?  In other words, is the Linux TCP/IP stack
(w/ Zebra/OSPF) be more flexible?  Miguel, I don't really expect you to
answer this part of the question because you're not supposed to be a Linux
expert; maybe someone else on the listserve would know.  But, if you do ...

Thanks,
Dennis

Reply via email to