ColdFusion Muse

Multiple Garbage Collectors: Can Two Sanitation Engineers be Better Than One?

I ran across this JVM configuration on a server recently. I should note that the server in question was having some issues, so this is not an endorsement of this approach. I simply had not seen this sort of configuration on a Coldfusion server before. Here are the JVM arguments:

java.args=-server -Xms512m -Xmx768m -Dsun.io.useCanonCaches=false -XX:MaxPermSize=192m -XX:PermSize=64m -XX:NewSize=48m -XX:+UseParNewGC -XX:+CMSParallelRemarkEnabled -XX:+UseConcMarkSweepGC ...
What seemed unusual to me is that this particular set of arguments allows for 2 garbage collectors to be specified. Both the "UseParNewGC" switch and the "UseConcMarkSweepGC" switch are set. I did not know that multiple GCs could be specified. I set about finding out how this was possible.

If you have ever used the Coldfusion Administrator to change garbage collectors you may have run into the issue where Coldfusion adds the UseParallelGC garbage collector back into the arguments. For example, if you change the garbage collection to "UseConcMarkSweepGC" and then update (again - using the Coldfusion Administrator), you will end up with both the UseParallelGC and the UseConcMarkSweepGC as arguments to the JVM. When you restart Coldfusion it will fail - complaining that it cannot handle 2 separate GC arguments. See this previous post for a more thorough discussion of that issue.

However, this does not mean that all garbage collectors are mutually exclusive. In my reading I missed that tidbit. The arguments above show how they can be made to live together. After a bit of googling I found several warnings that explicitly tell me that "UseConcMarkSweepGC" and "UsParallelGC" are indeed mutually exclusive. Since that is the error I ran into I suppose I jumped to the conclusion that GC's are mutually exclusive in general. Now I know that this is not always the case. Note also the CMSParallelRemarkEnabled switch in the example above. This switch is designed to reduce the pauses that occur when the GC is marking objects for deletion.

In my searching on the topic I found this document in the help files for CAMS (a network security policy server from Cafesoft which runs on Java). While some of the advice is specific to CAMS, I found the overview to be excellent. In particular there is some good information about the differences between multiprocessors and single processor machines - and why various garbage collectors work or do not work well with them.

I know there are some Java/Coldfusion gurus reading my blog. Perhaps you could chime in with some additional information.

Comments
Sami Hoda's Gravatar Good post... it leads me some additional tweaks I can do on my systems.

We're always looking to find the best "Sanitation Engineers"...
# Posted By Sami Hoda | 11/29/07 2:00 PM
rish's Gravatar It's good to take note that the ColdFusion Administrator can actually "work against" you when you're dorking around with -XX switches. (It is essential to remember that -XX switches are JVM implementation specific. Depending upon which JVM vendor you're using, switches that you read about in one place might not be valid with your specific JVM.)<br/>
Garbage collection has gotten to be a bit more esoteric with all the different algorithms being utilized by different vendors. A good over view of Java garbage collectors can be read about in a chapter posted online from the book Inside the Java Virtual Machine (http://www.artima.com/insidejvm/ed2/gc.html), by Bill Venners. It is important to note that if you read about garbage collection a few years ago you might want to brush up, some things have changed. While I would never suggest skipping the real meat of any matter, such as Venners’ aforementioned work, there is another article that provides some shorter explanations and graphs (http://www.softwaresecretweapons.com/jspwiki/thela...) by Pavel Simakov.<br/>
It’s also important to remember that certain design choices can actually trip up your garbage collector and Brian Goetz discusses them in an IBM developerWorks article (http://www.ibm.com/developerworks/library/j-jtp012...).<br/>
Finally, while it has gotten a bit dated now, there is a presentation by Nagendra Nagarajayya (http://web.princeton.edu/sites/isapps/jasig/2004su...), that while very verbose, highlights some important tools (towards the end of the presentation) that would be helpful for evaluating an adjustment to the garbage collectors; jconsole and jstat.<br/>
# Posted By rish | 11/30/07 9:01 AM



Blog provided and hosted by CF Webtools. Blog Sofware by Ray Camden.