One Domain; Multiple IPs.

[Barry Margolin]

In article <9j56fq$if7 at pub3.rc.vix.com>,
Brad Knowles  <brad.knowles at skynet.be> wrote:
>
>At 3:05 PM +0000 7/18/01, Barry Margolin wrote:
>
>>  Lots of smart folks came up with these techniques.  What do you suggest to
>>  replace it?  If all the servers are in the same data center you can use a
>>  load balancing reverse proxy, but that's not a reasonable solution for
>>  geographically diverse servers.
>
>	I have stated my preference many times for using routing 
>techniques (anycasting, whatever) to find the closest server farm 
>that serves a particular IP address, and then using load-balancing 
>switches on that IP address to distribute the load locally.

I don't think routing-based techniques are appropriate for TCP-based
services, or anything that requires that an address correspond to the same
physical server over a long period of time.  We use it successfully for our
caching DNS servers, since it works quite well with DNS's UDP-based
query/response structure.  But I would not consider using it for a web
server address; a backbone routing change could suddenly redirect packets
to a different server, and lots of connection resets would result.

I agree that load balancing switches work well within the data centers.  We
use Distributed Directors for geographic distribution, and Local Directors
for local distribution.

I suppose if the DNS-based mechanism weren't available, something like HTTP
redirects could be used.  DD supports this, but since it's done in the
application layer it doesn't generalize well.  We also use DD for our mail
relay servers (we tell our customers to use outbound.sys.gtei.net as their
smart host, and the DD gives them the address of the closest one), and SMTP
doesn't have anything analogous to redirect.

I agree with your general sentiment.  It's best not to use a screwdriver
as a hammer, but if all you have is a screwdriver, you make do.

Furthermore, most of your complaints about devices like Distributed
Director don't seem to be related to the way they give out different
answers to a query, but are about the fact that they aren't a complete
implementation of DNS.  Just because a particular DNS appliance is
half-assed should not be an indictment of the general principle.  The web
hosting company that GTE acquired a few years ago at the same time as they
acquired BBN (which, coincidentally, was named Genuity, the name we adopted
when GTE spun us off last year) had a load balancing system based on a
patched version of BIND, so it did all the normal DNS stuff normally.
Would you have as much of a complaint about this type of solution?
Admittedly, a problem with this type of solution is keeping the patched
version in sync with the changes being made in the on-going BIND
development.

-- 
Barry Margolin, barmar at genuity.net
Genuity, Burlington, MA
*** DON'T SEND TECHNICAL QUESTIONS DIRECTLY TO ME, post them to newsgroups.
Please DON'T copy followups to me -- I'll assume it wasn't posted to the group.