My RSS Feeds
Yet Another Language Comparison
Sheesh. I'm reading the article and thinking, "This is almost exactly the sort of thing I really didn't want to see in JDJ." It reads very sensationally - a lot like the "Will BEA survive against IBM?" crap did. Nothing against the author, but it sounds like he has an axe to grind and he's working on it as hard as he can.
Kirk Pepperdine is dead on. For one thing... YET ANOTHER language comparison. Secondly, the test, no matter how micro-accurate, simply cannot be macro-accurate. Thirdly... the "interview" tone of the article is HIGHLY annoying. I want the data, at the very least, not a report of the data. Lastly, with some of his statements, I question the conclusions he comes to.
For example, the one running the tests says things got faster during runtime with repeated runs. With Java, that's expected and understood - a known factor. He also says C++ did it. Okay... but why? C++ is early-bound, so it doesn't normally modify the code at runtime like Java does. So why the improvement? Why no explanation? Why needlessly create doubt?
Urrrrgh.
I don't understand why you expect me to know why running a C++ program more
than once improves performance. I don't develop C++ compilers, runtimes, or
the Linux kernel.
However, I assume the kernel's disk cache made subsequent runs faster, plus maybe the kernel kept a space of memory just the right size open after the first time.
Keith Lea [keith@cs.oswego.edu]
Keith: it's called "part of research." You have an observation, and you
just let it hang there. No explanation, no hypothesis, just an observation.
It invalidates the whole concept.
Why would disk cache make subsequent runs faster? Where did you see the improvement - when it loaded shared libraries? Wouldn't the operating environment already have loaded common shared libraries?
Why would memory matter? Again, the kernel doesn't preserve RSS or DSS between runs (hopefully!) so that can be eliminated... perhaps CSS is preserved, a feature Windows has as well as most UNIXen, but that wouldn't affect allocation times.