FijiVM: Compile Java source to Native with Realtime GC

(Search didn't turn up this posted on LtU yet.) Since I want to be able to write video games in a high-level, cross-platform manner, I am always happy to learn about things like the FijiVM which purports to give me realtime GC for my Java code, and hopefully have to neither (a) pay lot$ for it [edit: oh it looks closed and requiring $ after all :-] nor (b) run it on big server iron.

The key part of the FijiVM is a set of garbage collector algorithms that offer various performance characteristics and real-time guarantees. You can find more information about FijiVM's GC in Schism: Fragmentation-Tolerant Real-Time Garbage Collection. PLDI'10.

The key insight to getting constant-time heap access is to replace tries with arraylets. Arraylets represent arrays as a spine (a contiguous array) containing pointers to a set of array fragments. While objects can also be represented as arraylets, the Schism ap- proach follows Siebert [18] and allocates these as linked lists of fixed-size fragments. This choice is motivated by the fact that most objects are small and are always of statically known size. Thus even with a linked representation it is possible to have a precise cost of any field access, and this cost does not vary.

SCHISM/CMR is the only collector that supports all the features; other collectors either can not deal with fragmentation or increasing the cost of heap accesses.

Please tell me if you know about other systems that offer portable, high-level, and performant environments. [Edit: There are actually plenty of JVMs that go for the RTSJ type stuff, and most appear to be commercial.]

(P.S. I liked the image that shows how complex a decent compiler can be quite a bit.)

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

Games and high level languages?

That's going to be the title of my phd thesis!!!!

I've had a great experience ( with F#, both under Windows and OS X (with Mono)...

'arraylet' is not burdened with patents I hope

The arraylet idea looks awfully like the system I planned to use for dodo (and a variation on finger trees.)

I don't think so

Arraylets have appeared in several other real-time GCs and GCs for compressing heaps, going back to at least 2000. Have a look at the related work section in the Z-rays paper .