Thank you for all your inputs and escuse me for the late summary.
I checked my webservers for the following parameters and configurations.
The problem was actually with the vignette application server that was
causing the slow performance. After tuning the application server,
everything looks fine.
Following are the th replies:
On Sun, 12 Mar 2000, Mike Ghicas wrote:
> Okay, first things first....
> Solaris Setup -
> Are all patches applied? What version of Solaris are you using?
> Are all of the duplex settings for the NIC card hard set (do not leave
> anything as autodetect or autosens - it usually gets messed up)
> Add the following to /etc/system
> *Netscape Server Stuff
> set maxusers=2048
> set rlim_fd_max=1024
> set rlim_fd_cur=1024
> I actually have the values for fd_max and fd_ cur set to double - but I think
> that 1024 is the limit.. try uping it....
> Netscape Server Config:
> Are you looking up hostnames? This slows stuff down.
> magnus.conf (these are my settings YMMV)
> RqThrottle 512
> ListenQ 1024
> Have you double checked the local director?
> Do you have maxconns set in the local director config? This will limit the
> number of connections to a "real" server
> Do you get the same results if you hit the web servers directly oronly when
> going through the localdirector?
> Also, i had a friend tell me about slowness using story server - but i don't
> remeber the details... Check the benchmarks one piece at a time -
> w/storyserver w/o story server etc........
> -Mike Ghicas
Some of them pointed me to goo links for webserver performance tuning.
follow the link for a search on www.sunworld.com for sizing,securing
Solaris TUning links:
SOlaris -2.x- tuning ur TCP/IP and more
NES 3.X Tuning guide for solaris
> Venu M Middela wrote:
> > Hi Gurus,
> > I have a question for you all.
> > Our web site is going live very soon. we are in the Performance
> > and stress testing stage.
> > The current setup is something like:
> > we have 5 Sun boxes,1 cisco localdirector.
> > 2 boxes run NES3.63 web servers(E 250s each nodenames web1 and web2)
> > and 2 boxes run Vignette Storyserver applications adn other
> > applications run on these boxes(app1 and app2 E450s
> > each) and a oracle8i db running on a E4500.
> > The current scenario is that the cisco LocalDirector load balances the web
> > traffic between web1 and web2. All of the boxes are conneted to a HP x256
> > SAN.
> > recently we did a performance and stress testing on all of these boxes.
> > When we run the performance testing using some scripts that record
> > the highest number of hits and the time to retreive the pages and the
> > response time etc. we found that when the load on the either of the
> > webservers web1 and web2 touches 250 concurrent users per session, the
> > web servers can not take any more load and the web servers are down.
> > I have increased the variable RqThrottle in magnus.conf from 512 to 1024,
> > but without any success. the performance is still the same.
> > The way the applications work are: the user comes to the webserver and the
> > user request is forwarded to the application server and in turn forwarded
> > to the database and the db responds back the response to app. server and
> > the app server responds back to the web server.
> > Now, my question
> > is: what are the parameters and its max values that I need to set up
> > using ndd to increase the performance or to make the webservers bear more
> > load. Are there any other issues I need to take a consider?
> > The current values are:
> > tcp_close_wait_interval 240000
> > tcp_keepalive_interval 7200000
> > tcp_rexmit_interval_min 200
> > tcp_conn_req_max_q 128
> > tcp_conn_req_max_q0 1024
> > tcp_slow_start_initial 2
> > Thanks a lot and I will summarise,
> > Venu.
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:14:05 CDT