My original query :
Hi Managers !!!
Hardware : Sun Ultra 170E with 256 Mb RAM Solaris 2.5.1
I am performance testing fast cgi applications on this platform and am
noticing several TIME_WAIT's from the netstat -a on sockets that have
closed by the application. What parameter should I set with ndd /dev/tcp
in order to enable the kernel to "clean out" these connections and what
repercussions can I expect to have when doing this i.e. what downside to
the solution so to speak ????
NOTE: The performance test simulates several users hitting the Open
Market HTTP Server which handles the fcgi requests.
The responses :
From: Casper Dik <casper@holland.Sun.COM>
This is how TCP works; conenctions are supposed to linger in TIME_WAIT for
a period of time (4 minutes, I believe).
Unfortunately, with 100 hits/second that gives you 24000 TIME_WAIT
which take memory and time to process.
Sun now sells a special version of 2.5.1 which is much more efficient in
processing TIME_WAIT connections and connection set up. The special version
of 2.5.1 contains code that will be standard in 2.6.
From: Frank Pardo <firstname.lastname@example.org>
The TIME_WAIT status is part of the TCP protocol. It's a safety feature,
to prevent "stray" packets from corrupting later connections. You can
read about it in any good networking textbook, such as W. Richard
Stevens, "Unix Network Programming". Ignore it. After a short time, the
connection in TIME_WAIT state will shut down.
In summary, from the repsonses I received, I guess I will have to wait
for Solaris 2.6 to make its rounds.....!!!!
Rahul Roy Voice: (609) 727-4600 Ext.2650
Systems Consultant Fax: (609) 727-1318
Professional Services Group Mobile: (908) 202-5029
Bluestone Consulting Inc. Pager: (800) 631-4621
1000 Briggs Road Email: email@example.com
Mt. Laurel, NJ 08054 Web: http://www.bluestone.com/people/roy.html
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:11:11 CDT