DNS causing performance issues

From: Dan Norris <dannorris_at_dannorris.com>
Date: Mon, 17 Mar 2008 15:45:15 -0400
Message-ID: <47DECA4B.7040009@dannorris.com>

I'm unable to get a lot of information about the issue reported to me earlier today, but from what I can discern, there is an extract process that runs from sqlplus on serverA and connects to a database on serverB. The sqlplus session runs a number of pl/sql packages to generate CSV files that are spooled using sqlplus (on serverA).

The symptom reported to me was that the extract processes was slow. It usually takes about 45 minutes and ran for about 5 hours (it did complete successfully). Someone suggested that a previous performance problem was resolved by ensuring that proper reverse DNS records were in place. In this case, they found that those DNS records did not exist--nor did the forward records, but there were a proper /etc/hosts entries on the servers involved. This process runs once a day. After the performance issue, the DNS was updated to have proper forward and reverse DNS records. The next day, the extracts ran in about 32 minutes--faster than before. I'm chalking it up to coincidence.

Have any of you seen DNS records make a performance difference for sqlplus operations? I've seen connect-time delays for telnet, ssh, and ftp caused by DNS boo-boos, but not sqlplus. Even so, after connect-time, DNS isn't a part of the picture, so there's no reason for DNS to have an impact on performance after the connection is established. Just looking for any feasible theory on how these two facts can be connected.


Received on Mon Mar 17 2008 - 14:45:15 CDT

Original text of this message