Number of web servers

We are in the process of upgrading to 8.12 with TR 8.96.1

Just wondering how many web/jas servers others are using?

What rule of thumb do you use for # of users peer server ?

Do you share the load by using DNS round robin ? Any problems ?

Thanks in anticipation

[ QUOTE ]
We are in the process of upgrading to 8.12 with TR 8.96.1

Just wondering how many web/jas servers others are using?

What rule of thumb do you use for # of users peer server ?

Do you share the load by using DNS round robin ? Any problems ?

Thanks in anticipation

[/ QUOTE ]


I configure the web side with:

1- Production Isolation - A webserver for non-production and webserver(s) for production
2- DNS round-robin for budget conscious customers, hardware load-balancing for those with deep pockets.
3- Multiple production servers with combined HTTP/WAS using both vertical and horizontal WAS cloning to ensure failover and performance load-balancing.
4- Name-based virtual hosting (http://www.thestevensons.org/node/view/647)
5- Network Deployment on E1 Deployment Server

Some others may be able to give you # of users per server but the old rule of thumb is 250 users per jvm and one jvm per CPU (real CPU, multicore CPU but not hyperthreaded). Thus, if you have 4 dual-core CPU's you would create 8 jvms using vertical cloning and have (theoretically) capacity for 2,000 users, assuming an appropriate amount of RAM. If you have a second production webserver (which you should) with the same 4 dual-core CPU's you would horizontally clone the original jvm another 8 times and have (theoretically) capacity for 2,000 more users. I'm sure one of the other guys will mention that these are probably unrealistic numbers but it demonstrates the power of cloning.

Oh yeah, you'll hear many theories on jvm heap size. The current, popular number is 1024M min/max. I like many 768M min/max jvm's.

See below

_____

From: [email protected] [mailto:[email protected]] On
Behalf Of brother_of_karamazov
Sent: Wednesday, July 11, 2007 8:57 AM
To: [email protected]
Subject: Re: Number of web servers



Quote:
_____


We are in the process of upgrading to 8.12 with TR 8.96.1

Just wondering how many web/jas servers others are using?
We have two production servers (2 CPU/machine) clustered horizontally and
vertically for a total of 4 JVM's We also have 1 box for non-production
environments and 1 instance for JPD812 DataBrowser. This way, if
DataBrowser crashes the system it is on a non-production box.

What rule of thumb do you use for # of users peer server ?
Oracle still recommends 1 JVM per CPU with up to 150 users per instance/JVM


Do you share the load by using DNS round robin ? Any problems ?
I configured the Network Deployment (deployment manager) that has round
robin. We have not had any problems so far, but our concurrent user
population in only around 150 total users.

Thanks in anticipation


_____





I configure the web side with:

1- Production Isolation - A webserver for non-production and webserver(s)
for production
2- DNS round-robin for budget conscious customers, hardware load-balancing
for those with deep pockets.
3- Multiple production servers with combined HTTP/WAS using both vertical
and horizontal WAS cloning to ensure failover and performance
load-balancing.
4- Name-based virtual hosting (http://www.thestevensons.org/node/view/647)5-
Network Deployment on E1 Deployment Server

Some others may be able to give you # of users per server but the old rule
of thumb is 250 users per jvm and one jvm per CPU (real CPU, multicore CPU
but not hyperthreaded). Thus, if you have 4 dual-core CPU's you would create
8 jvms using vertical cloning and have (theoretically) capacity for 2,000
users, assuming an appropriate amount of RAM. If you have a second
production webserver (which you should) with the same 4 dual-core CPU's you
would horizontally clone the original jvm another 8 times and have
(theoretically) capacity for 2,000 more users. I'm sure one of the other
guys will mention that these are probably unrealistic numbers but it
demonstrates the power of cloning.

Oh yeah, you'll hear many theories on jvm heap size. The current, popular
number is 1024M min/max. I like many 768M min/max jvm's.

Brother Of Karamazov/Jeff Stevenson! <http://www.thestevensons.org>



_____


The entire <http://www.jdelist.com/ubb/showflat.php?Cat=&Board=> JDELIST
thread is available for viewing.


Looking for a job? Check out the Job


This is the JDELIST EnterpriseOne Mailing List.
JDELIST is not affiliated with JDEdwardsR.

To unsubscribe from this list via email, Click <mailto:
[email protected]?Subject=Unsubscribe&Body=Sirs,

Please remove
this address from the JDELIST EnterpriseOne Mailing List.

Thank You.>
here.

If you have 4 dual cores and you put up 8 JVM's, what about the operating system. We always leave 1.5 to 2.0 gig for the operating system. In our environment, we only put up 200 users per JVM. Alot of factors come into play; thread pool size, timeout settings for BSFN's and grids (basically the JDBC pipe to the enterprise server) and what applications users are executing. An application with a large grid that maintains a two hour time limit with 250 users could cripple your system. We have 4 web servers with 3 JVMs on each one, we have limited the user timeout to 2 hours, the BSFN timeout to 5 minutes, and the grid timeout to 30 minutes. Our servers our load balanced using a Cisco router. The JVM's are load balanced using a custome JSP dispatcher page.

[ QUOTE ]
If you have 4 dual cores and you put up 8 JVM's, what about the operating system. We always leave 1.5 to 2.0 gig for the operating system.

[/ QUOTE ]

Not sure if you are referring to CPU capacity or memory capacity. For CPU, I use up the cores with JVM's and count on CPU scheduling to balance the CPU load requirements between the JVM's and the OS. As for memory, I configure the JVM heap sizes to leave somewhere around 768 or so for the OS.

[ QUOTE ]


I configure the web side with:

1- Production Isolation - A webserver for non-production and webserver(s) for production
2- DNS round-robin for budget conscious customers, hardware load-balancing for those with deep pockets.
3- Multiple production servers with combined HTTP/WAS using both vertical and horizontal WAS cloning to ensure failover and performance load-balancing.
4- Name-based virtual hosting (http://www.thestevensons.org/node/view/647)
5- Network Deployment on E1 Deployment Server

Some others may be able to give you # of users per server but the old rule of thumb is 250 users per jvm and one jvm per CPU (real CPU, multicore CPU but not hyperthreaded). Thus, if you have 4 dual-core CPU's you would create 8 jvms using vertical cloning and have (theoretically) capacity for 2,000 users, assuming an appropriate amount of RAM. If you have a second production webserver (which you should) with the same 4 dual-core CPU's you would horizontally clone the original jvm another 8 times and have (theoretically) capacity for 2,000 more users. I'm sure one of the other guys will mention that these are probably unrealistic numbers but it demonstrates the power of cloning.

Oh yeah, you'll hear many theories on jvm heap size. The current, popular number is 1024M min/max. I like many 768M min/max jvm's.

[/ QUOTE ]

The general rule of thumb is about 1.5 MB minimum per html client user (Oracle's stated recommendation, not mine). Obviously, this is very small and in fact assumes that the heap size is set relatively low as well. I think a more reasonable number would be about 3 MB minimum per user. I am with Jeff, I always prefer to set the heap size to about 768 MB unless there are custom processes that will be used by the html client to load large amounts of data into the database. If this is case, then a 1024 MB heap size may be best as anything lower could lead to unexpected javacore heap dumps.

Hi guys,

I meant not to hijack this thread but how does one change the JVM heap size and where? I seem blind not to see it at the OAS installation guide...

thanks