[c-nsp] OT: Browser adoption of srv records

[Tim DeVries]

A few of years ago I was working on a geographically distributed ASP
type implimentation.  Requirements in order to avoid SQL merge
replication and to facilitate HA was a hot standby.  At the time we were
using CSS capable devices to loadbalance the configuration between
datacenters, but there was some amount of scripting going on in order to
make the N tier app infrastructure work.  Using Bind 9.x (as the
authoratative zone) we could specify RR sets of a pair of servers which
would allow delivery of ordered pairs of dns records (which would then
get quickly and multifariously scrambled by any third party dns in the
path).  The advantage of having an ordered dns set pair (or three) given
as an answer should be fairly straight forward.  My understanding from
the period was that IE caches dns for 15 minutes.  In a DR situation in
which one DC goes away, using the CSS or its equivalent for delivery of
dns.. results in the browser giving 404 due to the fact that it has
cached *one entry.  Even if the CSS or its equivalent was to give an
anwser of two servers, the order would get scrambled by third party dns
servers in the path.  Whereas an ordered pair would allow a user to
click 'next' in the app after a DC failed, and the browser would
automagically try the other server.  Of course if you don't care about
redesigning your entire application to support merge replication you
could just scramble the dns response or use proximity dns.

If I am incorrect here, please correct me.  It has been a few years.

Anyways, the idea I am faced with is why browsers do not support SRV
records.  If they did ordered sets would not be an issue, and port
information could be specified along with the dns by the click of
www.myurlhere.com.

Regards, 

Tim