[openib-general] Lustre Network Driver - KDAPL or verbs?

[Roland Dreier]

    > The driver we plan to develop should strive to address several goals: 
    >  - high reliability and performance

It seems unlikely that you would get more reliability or performance
by adding another layer of software in your stack.

    >  - allow interoperability between user and kernel level
    >  - allow interoperability, or better, portability among different
    > operating systems (Linux, OS X, Windows, Solaris)

Interoperability seems a function of designing an appropriate wire
protocol rather than how you choose to implement the protocol.

I believe that experience has proven that trying to maintain a single
codebase portable to different OS kernels is always more work than
just having separate codebases for separate kernels.  Even trying to
use the same code in both Linux kernel 2.4 and kernel 2.6 is enough of
a pain that it's probably not worth it.

    >  - be suitable for inclusion in the Linux kernel

It extremely unlikely that kDAPL will ever be included in the kernel.

Does this last point mean that you are planning to try again and work
on merging Lustre into the mainline kernel?

 - R.