Optimum Max JVM Heap Size Question

[MichaelSJudd]

I saw Ken running around, so I figured this would be a good time to post.

I am running CF 6.1, Windows 2003 Server, IIS 6.0, SQL2003; all are on the
same machine.

This machine is a web-server that runs this particular ColdFusion application
ONLY. Nothing else.

I have looked through a decent amount (although I'm sure not all) of the white
papers on this setting. I have left it at 512MB, but since I have 2GB on the
server, I thought about increasing it, but didn't know how much CF would really
use, if it would increase garbage collection times, etc.

So once my brain finally said "I don't know", I thought I'd post on it.

Thanks,

Mike

[ke4pym]

Tuning ColdFusion is an art. Generally I set my Xmx to 1024 and my Xms to 384.
Simply increasing the JVM heap isn't going to cause the GC's to take
significantly longer as long as you keep their size in check as well.

Are you having issues now with stability and out of memory errors?

[MichaelSJudd]

Yeah, I gathered it was an art.

Stability/performance has been a minor factor - the SQL database referenced by
the datasource is on the same box, and I still get the occassional "datasource
went by-by" (i.e. - timeout on dsn) error. It seems to occur during peak load
times (peak query load, that is). Disk activity goes pretty high at those
times, as well.

I was kind of going with the "better from memory than from disk" idea, but
didn't know if this was a dance with other settings (i.e. - max queries, etc.).

Sounds like it is - so I will start (gently) changing a few parameters to see
if I can get those DSN lost errors to be extremely infrequent.

I really appreciate the reply.

- Mike

[ke4pym]

You might also want to tinker with limiting the number of threads each DS can
create to the database as well as tinkering with maintain connections.
Generally, we limit our DS to 12 threads and do not maintain connection.
Unless it is a crazy busy DB. Like, say, the DB for client storage data.

[MichaelSJudd]

I'll try that - those are great ideas!

The software is live, but still considered beta for our customers, so I have a little bit of leeway.

I really appreciate your help!

- Mike

[cgsj_usa@yahoo.com]

Why wouldn't you want to maintain connections? Isn't it better to use the same connection rather than to spend resources creating a new connection? Thanks.

Chris

[MichaelSJudd]

Chris,

Personally, I DO cache the connection - as ke4pym mentioned, you wouldn't want
to turn it off for the client storage data DSN.

The idea seemed to be more like - "You should probably at least think about
tuning these to meet your specific requirements."

It would also have an effect depending on whether you used pooling for the
driver in SQL, etc.

- Mike