Re: The Kommunity vs. Dick Johnson

Erik Corry (erik@arbat.com)
Wed, 18 Nov 1998 00:21:04 +0100


You need to make a few changes to the test setup to make
it fair.

* You have to move the C code to a different file, to stop
the C compiler doing interprocedure analysis and only
calling the checksum routine once. This works against
your version of the code of course.

* Your system with alarm() calls is flawed. For a 2
second sleep you are getting +- 0.5s actual time running.
You should probably use gettimeofday to test how long the
code really ran.

* When you use RDTSC it doesn't work on non-Pentiums and
it gives unreliable results on dual systems, where you
might accidentally switch processors between tim() calls.
You should at least document that.

Due to all these things, I am going to stop making any
sort of judgement on the code. The test harness is too
unreliable.

-- 
Erik Corry erik@arbat.com           Ceterum censeo, Microsoftem esse delendam!

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/