Compress with pbzip2

[ltworf]

I think that rather than having a compression module, the code could just run an external compression program and then read back from it.

This would allow us to use more diverse tools.

I think we should use pbzip2 for this, since db servers have many cores.

I run this experiment:

# asd is a file filled with random data, sized 149MiB
$ time (cat asd | pbzip2 > asd.bz2)

real    0m6.155s
user    0m40.856s
sys     0m0.660s

salvo@vulcano /tmp$ time bzip2 asd 

real    0m21.327s
user    0m20.692s
sys     0m0.156s
salvo@vulcano /tmp$ 

As you can see pbzip is clearly faster, even on streamed input, not just on mappable files.
This would benefit, by reducing the backup time.

[ltworf]

It would also make the compress-test useless.