Re: bad lmbench numbers for mmap

Linus Torvalds (torvalds@transmeta.com)
Sun, 25 Apr 1999 23:32:24 -0700 (PDT)


On Sun, 25 Apr 1999, Larry McVoy wrote:
>
> So do you want me to change it to do reads instead?

Yes.

But save the old version away as a "lat_mmap_write" or something.

Btw, you need to fix lat_mmap anyway: right now the summary information
gives random numbers that depend on how large an area the user asked you
to use. So you can't compare lat_mmap by looking at the summary stuff,
which kind of defeats having a summary in the first place.

I'd suggest taking the 1MB (or some other fixed size) number into the
summary.

> Another thing I could do is to factor out the cost of mmap by running two
> experiments:
>
> a) repeatedly read a region which is already mapped
>
> b) repeatedly mmap & read & unmap.
>
> In theory, b - a == mmap+munmap cost. However, if the munmap causes the
> caches to get flushed, or if the new mapping is somewhere else, then cache
> misses start skewing the results.

I don't think it should much matter. But I don't have any numbers, so..

Linus

-
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/