Been doing a bit of performance testing this evening on the R GUI with RExecServer. Qualitatively, on my year old MacBook Pro, it seems that nothing has changed. At least, the overhead incurred by drawing the graphs or outputting the text seems to outweigh the communication overhead by quite a bit. This may change when hinting gets going, but thus far there doesn't appear to be a problem. Additionally, we have plenty of space for optimization in the distributed objects interface since we're not defining protocols yet. Good good good.
One thing does worry me: shipping bitmaps across the wire. I *think* that NSData it is actually optimized to use shared memory, but I'm not positive. In any case, the bitmap size of a full screen graph is pretty huge. That said, we could always do some timing (hey, we're statisticians, right? Staunch empiricists.) and then compress bitmaps above a certain area for transport via libz or something suitably fast. Hell, RLE would probably be a big win in most plots. I know you're thinking "use PDF," but PDF rendering performance is really awful relative to compositing to a bitmap. (Witness complex scatterplot in your Keynote presentation. There's a reason it takes 30 seconds for your graph to come up.)
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment