Thanks for the replies Chuck as always; I meant to say that sooner. =P
I am also sorry this reply is a week overdue, however I have been tied
up for a week.  Let me explain the situation a little more thoroughly
and see what your thoughts are then:

For several critical applications such as SSO and student registration,
we want them clustered over multiple servers behind a load balancer.
The SSO application's load consists of users logging in and ticket
processing.  The student registration application's load consists of
thousands of students pounding at a handful of servers constantly in
hopes they can schedule their classes before they fill up.  It is like
people watching an eBay auction and constantly refreshing the page to
see if they have the winning bid and upping it whenever they aren't.

If a server goes down, the load balancer will forward traffic to the
others such that no one will experience interruption of service.
Obviously, this requires the application servers to be clustered such
that their sessions are replicated with one another.


Given that our SSO solution is currently clustered among three machines:
 
1. Does Tomcat keep all session information strictly in RAM or does it
swap out to disk?  If it swaps out to disk, is there any way to
configure how much session information can be kept in RAM?

2. What commonly used formulas are there to determine sufficient RAM
based on load size?  I am currently reviewing logs from our peak
application usage as well as an average day to determine the number of
logins / second.

3. Is it better to split large clusters into smaller ones with a few
nodes replicating to the other clusters? (Example below with nodes
denoted by "O")

 SSO  CLUSTER
--------------
| CLUSTER  A |
|            |
|  O------O  |
|   \    /   |
|    O--O    |
|    |  |    |
|----|--|----|
|    |  |    |
|    O--O    |
|   /    \   |
|  O------O  |
|            |
| CLUSTER  B |
--------------


I posed my original question (Can Tomcat write session information to
disk or database like the Apache HTTP) because it seemed more scalable
as:

1. Our SAN has several terabytes worth of space so session information
can grow easily.

2. Reduces RAM bottleneck

3. Reduces network traffic to replicate session information


I would appreciate any thoughts and alternatives you have to offer!

Thanks,

Andrew R Feller, Analyst
University Information Systems
200 Fred Frey Building
Louisiana State University
Baton Rouge, LA, 70803
(225) 578-3737 (Office)
(225) 578-6400 (Fax)


-----Original Message-----
From: Caldarale, Charles R [mailto:[EMAIL PROTECTED] 
Sent: Wednesday, February 27, 2008 2:17 PM
To: Tomcat Users List
Subject: RE: Storing sessions to disk like Apache HTTP server

> From: Andrew R Feller [mailto:[EMAIL PROTECTED] 
> Subject: RE: Storing sessions to disk like Apache HTTP server
> 
> Actually, the system we are concerned about will have tens of 
> thousands of concurrent users.

Number of users != number of sessions, unless they all log in and stay
logged in concurrently (or you have a very high session timeout
setting).  Regardless, given the price of memory and the ready
availability of 64-bit platforms, there's no sense in over-engineering
this.

 - Chuck


THIS COMMUNICATION MAY CONTAIN CONFIDENTIAL AND/OR OTHERWISE PROPRIETARY
MATERIAL and is thus for use only by the intended recipient. If you
received this in error, please contact the sender and delete the e-mail
and its attachments from all computers.

---------------------------------------------------------------------
To start a new topic, e-mail: users@tomcat.apache.org
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]


---------------------------------------------------------------------
To start a new topic, e-mail: users@tomcat.apache.org
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to