On Mon, Feb 23, 2015 at 10:52:17AM -0500, John C Klensin wrote: > In that regard, it appears to be more like the single-service MX > record, which has never needed a weight. > > If "make this as much like SRV as possible" is a motivation, > then I'd expect that to be written a little more clearly and > might expect Weight to be specified as "reserved for future > extension; set to zero in all cases". Or, if not, I expect its > use to be clearly explained. Perhaps that explanation could be > just a reference to SRV, but I doubt it. A couple of years back I wrote (for my employer) a Python LDAP iterator that fully implements the SRV record priority and weight specification. I found that others looking over the code later were surprised by the attention to detail. It seems that many implementations don't even pay attention to SRV priorities, let alone weights. So perhaps you're right, in that the SRV priority+weight design is likely over-engineered in general, and for this use-case in particular. It seems likely that many implementations simply ignore the full semantics. -- Viktor.