Hacker News Reader

Thanks. I'll work a bit with unpacking all this. =)

Meanwhile. I just published an experiment where I have switched out the BitSet for an array of booleans. Just to demonstrate my observations that then the performance is perfectly stable. https://github.com/PEZ/ghost-chase-condition/tree/master/tes... Does that affect your analysis in any way?

View on HN

I don't know the JIT internals, but just ignoring them I'd point fingers on the setBit(x) = "load byte from mem into reg, set some bit in reg, write back reg to mem". Now it's just a "write byte to memory", which is much faster at the expense of using more storage (factor 32 or 64, probably 64). Though I am surprised caching doesn't mitigate this.

Reply on news.ycombinator.com