RE: Please read - proposed WG termination

--On 31. august 2005 20:10 -0400 Bernard King-Smith <[email protected]> 
wrote:

> Having IPoIB-CM is a very important feature to make IB a viable
> interconnect in clustered systems. Without IPoIB-CM for HPC clusters, and
> commercial clusters using IB to SAN, you need two networks for good total
> cluster performance, one for IB ( non-IP traffic ) and an IP performance
> network like GigE.  This means that IB is not cost effective as the GigE (
> or 10 GigE) network which handles both types of traffic reasonably.

Actually the numbers I've seen show that you can get more than 1 Gbit/sec 
(the test I was shown had 2 Gbit/sec) out of TCP/IP over Infiniband.
Still pitiful for an 8 Gbit/sec network, but faster than GigE.
(Limitation seems to be completely in the host stacks, btw)

So while parallell 10GigE and IB networks will probably give you more 
performance than an IB network alone, parallell 1GigE and IB networks will 
probably not.

Your mileage may vary, of course.

Apart from that, I agree totally with your logic - with the small provisos 
that 1) we do not know what the performance will be until we code it, and 
that 2) with smarter adapters and better stacks, performance of TCP/IP over 
DG is likely to improve too.

                          Harald

_______________________________________________
IPoverIB mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/ipoverib

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (MingW32)

iD8DBQFDFsX4OMj+2+WY0F4RAjJKAJ0edFYU0mneGMCB3/HiS870aDDWgwCg2yWr
AaVIwQbLTkH3Bbco+kt6860=
=Uca2
-----END PGP SIGNATURE-----