URLClassPath performance vs system loader

Question

I did some simple benchmarks on Sun 1.4 VM for Windows, and I noticed that the performance for URLClassLoader when loading classes on a local file system is approximately 30% slower than loading the same classes using the default application loader that the VM provides (i.e. as if the class was on the system class path)
I'm guessing this is because URLClassLoader is implemented (provided by the VM) entirely in Java and therefore has the overhead of the parser whereas the application class loader (also provided by the VM) is implemented natively. Or perhaps, this because URLClassLoader is limited by Java's file i/o performance. I haven't looked at the C src for JVM, so I'm not sure why.
I've also implemented my own file system loader (subclassed ClassLoader) and did the file i/o myself (i.e. using ZipFile, etc).
I was able to match URLClassLoader for performance, but could not get any faster. I don't think this is a surprise, since I'm probably unknowingly duplicating how URLClassLoader was implemented.
Can anyone offer any suggestions/insight on how to get URLClassLoader as fast as application ClassLoader for classes on the local/network file system?
A 30% slow down may not seem like a big deal, but it does become more noticable as the app becomes larger (more classes).

Answer 1

Can anyone offer any suggestions/insight on how to get
URLClassLoader as fast as application ClassLoader for
classes on the local/network file system?
You can speed up the io classes by writing your own, in some circumstances.
A 30% slow down may not seem like a big deal, but it
does become more noticable as the app becomes larger
(more classes).
You're right, it doesn't seem like a big deal to me. Exactly how often are you going to be loading classes? I wouldn't think the impact would be measurable unless you were loading a new class less than once a minute.