RE: Ecosystems cost of additional specs (RE: Please re ad - proposed WG termination)

> At 12:39 PM 9/3/2005, Carl Hensler wrote:
>
      > > Date: Fri, 02 Sep 2005 15:45:30 -0700
      > > From: Michael Krause <[email protected]>
      > > Subject: RE: Ecosystems cost of additional specs (RE: [Ipoverib]
      Please
      > re ad - proposed WG termination)
      > > To: Dror Goldenberg <[email protected]>, "'Harald Tveit
      Alvestrand'"
      > <[email protected]>
      > > Cc: [email protected]
      >
      > > Somewhat an aside, but it is ironic that standard APIs are often
      ignored /
      > > deemed "crap" by some even if many systems can benefit from them
      yet this
      > > spec which is likely to be implemented by a single OS and used
      only in
      > > perhaps a subset of deployments is deemed worthy. Apologies for
      the tangent.
      >
      > Mike,
      >
      > IPoIB-CM is an optional protocol extension accessed through the
      > standard networking API, Sockets.
      >
      > If standardized, it will be implemented for all three major OSes,
      > Linux and Windows in OpenIB, and Solaris at Sun.
>
> Didn't know Solaris was a major OS from a volume perspective (might be
said for any Unix these days as there really > > are only two OS of any
real volume any longer).  :)  I will state that at this point in time, I am
not recommending to my > > company to do any work on IPoIB-CM and I plan to
push on spending resources on where there is significant value and > ROI
opportunities on the RDMA side of things not on making IP-based stacks work
better (we are already getting more > than good enough performance with the
existing specs implemented).
>
> In any case, Windows is really a third-party provider doing the work not
MS.  Does not really matter I suppose in the > end.

I don't include Windows as a user of IB because it is pretty much a
Ethernet only OS looking at its current market.

>
      > I think the WG should complete work on IPoIB-CM expeditiously,
      > and determine whether there is community support for completing
      > the MIBs.
>
> I don't see the work as compelling but people are free to ignore me as
I'm just one voice within the IETF here.  I > personally believe the
emphasis should be on getting a solid RDMA infrastructure in place and
using that for ULP > communication than on making what is already a
reasonable IP over IB network stack more optimal.  If you go back > through
the history of interconnects, when one cannot demonstrate significant
benefit difference to customers or take too > long to get the value-add
right, then customers are fine with going with the evolutionary technology.
10 GbE RNIC prices > are dropping as are switch port costs.  While they may
not be on par with IB today, by the time it really matters to > customers,
the total solution costs which go well beyond the hardware may result in 10
GbE being just fine.

This is where technology and marketing get into trouble. If IB is not on
par, then 10 GigE is all people will use. Even though IB has other non-IP
uses where it excells, when you look at total cost, two networks are more
expensive than one. Hence, IPoIB just good enough means no IB for most
customers. I don't agree that 10 GigE can be everything to everyone, and IB
has a place. Make it cost effective, and they will use it. IP is not going
away in the foreseeable future, and many a person/company has proclaimed
the death of IP over some newer technology, only to end up in the memories
of time. The only networks that survived for any length of time supported
IP well (Token ring, FDDI, ATM), even if not always fully utilized by an
application set.

  10 GbE will > have LSO, LRO, whatever as well as RDMA and a much lower
cost to integrate management and associated ecosystems > into the data
center.  Spending resources on making IP communication better on IB is
simply a poor use of resources > with minimum end visible customer benefit
especially if some of the OS such as Windows already have the hooks in
place > to support RDMA ULP via Winsocks Direct.
>
> Mike_______________________________________________
> IPoverIB mailing list
> [email protected]
> https://www1.ietf.org/mailman/listinfo/ipoverib



Bernie King-Smith
IBM Corporation
Server Group
Cluster System Performance
[email protected]    (845)433-8483
Tie. 293-8483 or wombat2 on NOTES

"We are not responsible for the world we are born into, only for the world
we leave when we die.
So we have to accept what has gone before us and work to change the only
thing we can,
-- The Future." William Shatner