[develnytprof-dev: 650] Profiling parallel tests

[Nicholas Clark]

On Sun, Aug 17, 2008 at 07:40:35PM +0100, Nicholas Clark wrote:

> However, I'm pleased to report that for (inferred) highly repetitive data,
> such as running the parallel test harness, the compression is rather better.
> 
> $ ls -l nytprof.out
> -rw-r--r-- 1 nclark ma 64608399 Aug 17 19:03 nytprof.out
> $ ls -lh nytprof.out
> -rw-r--r-- 1 nclark ma 62M Aug 17 19:03 nytprof.out
> 
> 
> ie nowhere near close to filling the partition. WIN.

I added some code to append the raw and compressed byte counts, and the
compression ratio. In this case:

# Total uncompressed bytes 1654423529
# Total compressed bytes 77724458
# Compression ratio 1:21.285754, data shrunk by 95.30%


Mmm, there are another 1549 profile files, which (compressed) are about
another 50% on top of the main profile file, but I'm not sure how they
end up filling the 4G.

Nevermind. I like that number.

Nicholas Clark