hadoop-common-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Elliotte Harold" <erhar...@gmail.com>
Subject Re: Overhead of Java?
Date Fri, 14 Sep 2007 13:18:08 GMT
On 9/6/07, Ted Dunning <tdunning@veoh.com> wrote:
> These programs are not particularly compute bound in most cases and Java
> suffices to saturate the I/O bandwidth.
> Besides, reasonably written java code tends to be just as fast as similar
> C++ code.  Even for hard-core numerical applications, you have a very hard
> time beating java (take a look at the Colt benchmarks).

Java 6 has reached the point  where the real determing factor is whether a
particular problem and algorithm is more amenable to compile-time
optimization, in which case C++ is faster, or runtime optimization, in which
case Java is faster. Given the best available compilers and algorithms,
these days the difference is usually within a few percent one way or the
other, and so far I haven't been able to predict in advance which language
is likely be faster on any given problem without measuring it.

As Ted points out, for typical Hadoop style-applications disk I/O and
network latency and bandwidth issues are likely to make CPU considerations

Elliotte Rusty Harold

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message