3 ms·
from another discipline, but similar rigor -- a FOSS community with both C++ and Java servers, serving high-requirements responses.. a sort of friendly "coopeti
by mistrial9 6y ago
from another discipline, but similar rigor -- a FOSS community with both C++ and Java servers, serving high-requirements responses.. a sort of friendly "coopetition" contest around 2012.. many hands and eyes, certainly top-level engineering on both sides.
The results were basically -- with the best teams, almost identical times; the obviously slowest entries were on the C++ side. Admitting bias now, I was surprised to see Java winning by a "hair's breadth" in certain cases, with certain entries.
Later, talking to a PhD in CS who wrote their thesis on graph-analysis of code conditions for JIT optimization, I found out that Java put a lot of excellent computer science into the problem, and that works very well. As alluded to in a different comment, the response times in both camps were appproaching the physical limits of the hardware, apparently. I had to recalibrate my expectations, since I thought that C++ would certainly win without fail.
- AtlasBarfed 6y agoHow did they avoid GC penalty? Arena allocation and buffer reuse? Java is pretty good, I've used it for over 20 years, but it wastes memory for basic use cases, and yet GC pause limits the utilization of huge memory cases. Did anyone try Go? IMO Go tries to address the slowness of java startup and commands, but it punts GC/memory waste so the huge-memory systems are even worse.
- mistrial9 6y agothe results that counted were "warm" responses no one tried go-lang, the elaborate specialty code needed for the actual work did not exist the Java probably did use pre-allocation and flat heirarchy a lot, but I do not know the details. Some Java did win by a significant margin, in very odd corner cases, probably due to simple code-coverage miss on the C++ side