RE: [Ikvm-developers] Overhead and Best Practices

[Jeroen F. <je...@su...>]

Mark Proctor wrote:
> IKVM with Drools 2.x has been a great success. We are now=20
> looking to do something similar for Drools 3.0. However we
> are under pressure for a full port of the core code - which
> I want to resist due to the resources needed to maintain
> such an effort.
>=20
> One key issue that .Net developers bring up is the overhead of using=20
> IKVM - quoting figures of over 20%. I have no empiracal facts, but I=20
> find that hard to be true.

In general 20% sounds way too high, but some operations are
significantly slower. For example, exception handling is much slower
under IKVM than under either pure .NET or Java.

Dynamically loading Java classes is also pretty inefficient, but for
statically compiled code the overhead typically is low.

Another issue plays more at the design level, if you rewrite and
re-design your code for .NET you're more inclined to use .NET specific
features (like for example value types) that can affect performance in a
more significant way.

> Also maybe a "Get the Facts" document might help, as i have=20
> some users worred about a Frankenstein system; even though I tell
> them that running Java on .Net is no different than runny Python
> or VB, it all compiles down to IL.

That's true, but a significant drawback of IKVM is the fact that you
have to ship the Java class libraries (in the form of
IKVM.GNU.Classpath.dll).

Regards,
Jeroen