[openib-general] Suggested components to support in 1.0

[Bryan O'Sullivan]

On Sat, 2006-02-25 at 20:05 -0800, Sujal Das wrote:

> - OS distro and kernel versions to be tested with - RHAs, SLES, SUSE
> Pro, Fedora etc..., kernel 2.6.13, 14.... etc

We'll have to see how many of those are feasible.  Packaging userspace
for the different distros is easy enough; the big problem is backporting
kernel support to kernels that will actually work on the distros in
question, and building binary packages of those.

I think the only feasible approach will be for people to build binary
packages of whatever kernels they can and make them available.  People
who test can either use those, or build their own kernels and report the
kernel versions they are using when they send in test results.

> - CPU and chipset platforms

I know there's a need for good coverage on the following:

      * x86_64
      * i386
      * powerpc (aka ppc64)

I assume ia64 needs to be included, too, but I would very much like
people to let me know what platforms they want to see tested or can test
themselves.

> - Variations on each HCA hardware.  For example - with Mellanox, we
> have DDR, SDR, Mem-free etc.

I don't have a comprehensive list of the HCAs vendors plan to test.
This would be very useful to have.

> - Testing with various switches available

Likewise :-)

> - Testing on large clusters - 128 nodes and beyond ...
> 
> - Testing for performance/features with ISV / user level apps apps -
> Fluent, Oracle, DB2, LSDyna, MPI etc

For publicly available software, there's obviously no problem with this.
At least some ISVs are known for placing restrictions on the publication
of performance results, so while I'd like to see numbers, please be
careful in what you choose to report.

> - Testing with available storage targets - both gateways and native,
> SRP and iSER?

Yep.

It would be good to have a common set of features and applications that
vendors could test in a uniform way, so that we have at least a base set
of somewhat standardised test results.  In addition, any further testing
that people can perform and report on will be most welcome.

	<b