[AFS3-std] DNS SRV Resource Records for AFS

[Harald Barth]

> Per call is probably
> better in some sense, but I don't think it's sufficiently better to
> require implementations do it.

In what sense?

Debugging is easier if it doen NOT randomize for every call. Where are
the benefits for randomizing for every call? Can one client be such a
burden that it has to be randomized over serveral servers? It might even
be better for the servers if they answer to their (randomized) groups
of clients, but I have no numbers to prove one or the other theory.

Harald.