[rds-devel] RDS upstream

[Andy Grover]

RDS is now queued in net-next to go into 2.6.30. How is this going to
affect RDS and its development?

Not in a huge way immediately, but in a couple of important ways in the
mid- to long-term.

First, now that we know what RDS's address family number is -- 21. There
undoubtedly will be a considerable period where user libraries will need
to continue to support the current method of determining this via /proc,
but at some point in the future this can be dropped, which will be nice.

Second, having RDS in mainline should make it more likely that
distributions will pick it up, even before they actually move to a
kernel that includes it. It's no longer out-of-tree code, it will be
just one more thing to backport for them.

Third, it's another kernel tree for me to keep up to date :) Given the
long delays between a kernel release and its enterprise adoption, OFED
will remain RDS's primary distribution mechanism.

Fourth, we should expect more people trying RDS, and perhaps more
attention from random developers. This will especially be true if we
reintroduce a transport that does not rely on expensive hardware.
Someone already mentioned to me -- NFS over RDS? Hey, why not? Send
patches when you find bugs :) Reliable datagrams can be nice for a lot
of things. One concern might be that other users might want features
that stray from RDS's original goals. I think the existence of another
reliable-dg protocol -- SCTP -- means that RDS will be able to stay
focused on what it does best.

All in all, this is a big, positive step for RDS and its adoption.

Regards -- Andy