Re: split the load?

> Perhaps this could be resolved by checking to see what the RFCs say
> about DNS round robin replies.

I'm not convinced.

> My guess is that's only a "SHOULD" too, though if it's a "MUST".....

I can't see where you are going ...

> note that I'm more inclined to agree with Philip, since if you really
> want "random" contact to your MX hosts

... that's the only interpretation I can put on equal valued MX values ...

> then you can easily install a modern DNS implementation that *does* do round
> robin serving,

I disagree.
Not every MTA manager can dictate how all the NSs for their zone work.
Also, simply having the primaries for a zone running a RoundRobin NS isn't
very efficient at randomising at the fine level. DNS RRs are cached, so if
the NS used by the remote MTA doesn't RoundRobin, then the MTA will keep
using the same host first until the TTL expires. Forcing people to set a low
TTL isn't acceptable ....

> why burden exim with wasted code and effort

This has not been mentioned before as a reason for not doing it.

> that's more likely to undo round-robin efforts than anything else.

... that's what people have been plugging ...
However, I am not convinced -- why should it *undo* it ?
Using a RoundRobin NS will not guarentee that each time your MTA tries to send
to a service that you will get the MX's "round robin" -- you are a *lot* more
likley to get them "at random". As such, exim randomising will *not* undo the
effects of the NS. random(ramdom) is random !

1) whatever the RFCs say, for the forseeable future not all NSs will RoundRobin
2) randomising random ordering gives random ordering
3) randomising non random ordering gives random ordering
4) I would imagine that the cost of exim doing it is not large.

If (4) is wrong, let's come out into the open, and say it's too expensive !