SRV is great if only it were implemented

[Cricket Liu]

> Speaking as a "corporate type", my take on SRV is that it's a nifty
> service-locator mechanism, but as a general load-balancing-and-redundancy
> mechanism, it leaves a lot to be desired. Plus the major paradigm shift
> at the application level -- mapping services to hosts instead of just
> hosts to addresses -- is a bitter pill to swallow.

Why is this such a bitter pill to swallow?  I personally like the idea of
specifying a service I want to access and asking DNS to find a server
that provides it.  It's certainly better than having to count on people
to name their FTP server ftp.acmebw.com.  And there's a clear
precedent in the MX record, which maps the SMTP service to the
hosts that provide it.

> If you want *real*, honest-to-goodness Dynamic Load Balancing, get a
> commercial product that performs it. If you're on a budget, do the old
> "configure rrset-orders on all the authoritative servers and lower the
> TTLs" trick to achieve an approximation to the real thing (at the expense
> of requiring more DNS server maintenance, and generating more DNS traffic
> because caching is defeated).

But why not implement a record like SRV that doesn't require people
to buy extra hardware or software, and that provides simple failover
and load distribution?

cricket

Acme Byte & Wire
cricket at acmebw.com
www.acmebw.com

Attend the next Internet Software Consortium/Acme Byte & Wire
DNS and BIND class!  See www.acmebw.com/training.htm for
the schedule and to register for upcoming classes.