tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Kevin Wang <>
Subject RE: Statistics for Tomcat
Date Wed, 02 Jan 2002 18:05:45 GMT
Performance will vary widely depending on your servlets and jsp's.  We need
more information before we could even begin to start giving you any
information.  For example:

Question: We have servlets that tend to run 5-10 seconds each call, and want
to support 500 concurrent requests at a time. what kind of architecture will
support this?

Answer: need more cpu usage information.  For example, the servlets may take
5-10 seconds, but may take up 0% cpu because it's waiting on some external
resource, like a database call.  Or, the servlets may take up 100% cpu.  If
the former case, then a single machine running tomcat would suffice.  if the
latter, then you would need 1 cpu per user!  

Also, you need to clarify between concurrent users and concurrent requests.
users are not 100% active.  Our model here at work is a user will click on
something once every 30-60 seconds, and be idle the rest of the time.
Therefore, a 10 second servlet being run every 30-60 seconds represents a
10/40-10/70 = 25-14% duty cycle of cpu usage.

Performance is a wily beast, and must be handled on a case-by-case basis,
but there are some other things that may be relevant.  For example, apache
in front of tomcat has some weird default limitations:

Apache by default (on unixes) is compiled for a maximum of 256 connections
at a time.  If you have servlets that are taking 10 seconds to complete,
this means you need to be very careful how many connections you can handle
at a time!  Given a set of users running a fixed 25% duty cycle (connections
are active for 10 out of every 40 seconds) that means that for 256
connections (assuming no connections for GIFs or other included content like
frames) Apache can handle a maximum of 256 / 25% = 1024 users.  

IMPORTANT: This is a huge generalization, and only specifies a maximum,
given that users are 100% consistent and there's no variability in system
response time (read: a unrealistic, performance lab, static load condition).
Real world performance will likely be lower, especially if you need
additional cpu power for SSL encryption!

   - Kevin

-----Original Message-----
From: Mrutyunjay []
Sent: Tuesday, December 18, 2001 1:47 AM
Subject: Statistics for Tomcat

I want to know where can I find the information regarding the performance
statistics of Tomcat, based on which I can decide upon what deployment
architecture to be employed for my project 

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message