Andrew Siwko's Blog #3

This is a test of w.bloggar 3.02 posting to roller on my site. It’s neat but I was hoping for something which would integrate the upload file and insert image functions into one process.

http://new.blogger.com/home.pyra was a great way to get started with web logging.  Roller’s installed and working here so I copied the content over (by hand one entry at a time preserving dates) and now it’s all here in the Stuff category.

Here’s a page from theinfocenter on how to set up JNDI caching for WebSphere 5: http://publib.boulder.ibm.com/infocenter/wasinfo/index.jsp?topic=/com.ibm.websphere.nd.doc/info/ae/ae/rnam_example_behavior.html

I’m not convinced about the wisdom of this: http://www-106.ibm.com/developerworks/java/library/j-ejb0924.html but there are several patterns which advocate caching of JNDI.  This would seem to be a pattern for container vendors not developers.

While working on an apparent MySQL connection hang I found this:  http://www.mysql.com/documentation/connector-j/index.html#id2802198.  So now I’ll be able to start thinking about an internet facing database!

I had a breakthrough today with my Tomcat hang. I’m using 5.0.14 beta on the site. It seems that from time to time Tomcat would hang up. On a whim I connected to the Tomcat manager app after roller and stuff hung up. It came up fine, there was no GC going on or anything. It smelled like database so I looled in the log and saw some io errors coming from the MySql driver. Here’s a link: http://www.mysql.com/documentation/connector-j/index.html#id2802490. It looks like autoReconnect is set in roller so there’s still a mystery.

The idea of a Java mail system is intriguing. I’d like to be able to have more control of my mail system. It’ll be hard to give up sendmail, though. http://james.apache.org

I could really benefit from spending some time working out a logging
strategy. Some output to syslog makes sense with the JavaCron code. Each
target class should have its own logger so they can run in parallel and
not lose or mingle output.

Weird things have been happening with belloc. I can’t correlate it
directly, but this machine has never worked for a long time without error
reading temperatures froma scanner. The other two boxes worked for a long
time. I am using IBM JDK 1.4 with Javacomm and this may be the source of
some of the trouble. I tried 1.4 on mary-grace but had to back out when
it didn’t work. The .available() method always returns 0. I’ve got a
listener registered now to see what’s really happening. There may be some
flakieness in the scanner also. I need to test it on another machine
before debugging this further.

You are given nine coins and a balance scale (you can accurately compare
the weight of two things). The coins are indistinguishable except by
weight. One is lighter than the other eight. What is the fewest number
of weighings you would require to identify the one coin.

A customer has a p690 with 32 processors and 32 Gig of RAM. Two of his
engineers are arguing over how many copies of the applicaiton server to
run on the machine. The application has proven to scale well but session
persistence has shown a small overhead when used with no session affinity.
There are no other applications running on this machine to compete for
resources. One engineer wants to create one big JVM with 32 Gig of RAM
assigned as max heap (assume this is possible, tell me why if it isnt).
The other wants to create 32 JVMs, one for each CPU.

What is your advice?
What are their arguments?