> On Jan 5, 2016, at 5:53 PM, Valdis.Kletnieks@xxxxxx wrote: > > On Tue, 05 Jan 2016 16:51:09 -0800, Joe Touch said: > >> This sounds like an opportunity for SGI to shift over to DNS-SD. >> >> We can change the line about "no known" to "only one known". We could >> refer to DNS-SD going to PS as a rationale for obsoleting TCPMUX if >> necessary, IMO. > > Not a really good fit. In general, you already *know* the hostname that > has the service - it's the host you have the NFS mount from, or the current > CXFS metadata server if you're inside the cluster. And trying to use DNS-SD > for two logically distinct DMF archives is just *looking* for trouble. You can use DNS-SD to just find the port number given s known hostname. The simplest alternative to that is to get an assigned (user) port number. > It would be a much better fit for portmapper actually (although that would > involve leaving a daemon running to answer/forward queries - probably not as > big an issue today as it was when they started coding DMF almost 25 years ago). The forwarding is why import mapper might not be a good fit. Joe