*Exactly* what is wrong in testing the good old way, by checksuming
a large datablock and measure the time it takes?! Answer: ** NOTHING **
I'll admit that I developed my Sparc checksum code from userland with
a test program that tested:
a) Whether it gave the correct calculations
b) How fast it was for various buffer sizes and alignment
However, after I was satisfied with what the program told me about my
code, I took the results very lightly until I saw what it translated
into when things are going over the wire with 2 or 3 interfaces maxed
out.
For writing and testing the initial code, sure, nothing is wrong with
these sorts of tests. But don't come to any final conclusions until
you stick the code in the kernel and measure it's effects with other
(I/O and cache) activity going on along side of it.
--------------------------------------------////
Yow! 10.49 MB/s remote host TCP bandwidth ////
over 100Mb/s ethernet. Beat that! ////
-----------------------------------------////__________ o
David S. Miller, davem@caip.rutgers.edu /_____________/ / // /_/ ><