bzip2smp 1.0 review

Download
by rbytes.net on

bzip2smp parallelizes the bzip2 compression process to achieve a near-linear performance increase on SMP machines. This program pa

License: BSD License
File size: 0K
Developer: Konstantin Isakov
0 stars award from rbytes.net

bzip2smp parallelizes the bzip2 compression process to achieve a near-linear performance increase on SMP machines.

This program parallelizes the BZIP2 compression process to achieve a near-linear performance increase on SMP machines. On a two-processor Xeon machine, the speedup is around 180%.

The tool's main purpose is to aid performing heavy-duty server backups. It can also be used on modern desktop multicore processors (AMD Athlon64 X2, Intel Pentium D etc).

There is NO speedup coming from hyperthreading on the hyperthreaded machines, since hyperthreads don't have dedicated caches, and the bzip2 is very cache-dependent. Expect degraded performance if you try utilizing hyperthreads.

The compression process requires more memory than the normal bzip2 -- some 15Mb average for 2 CPUs, 30Mb for 4 CPUs, etc. This should not pose any problem on a typical memory-rich server/workstation hardware, though.

The resulting archives are bit-by-bit identical to the ones produced by the normal bzip2, at least as of version 1.0.2.

No decompression is supported. The compression is stdin-to-stdout only.
If you need the missing features, you are welcome to implement them. Maybe someday the program will be fully interchangeable with bzip2, as a result. For now, it is not. Please also note that there is a similar program out there, pbzip2.

Unfortunately, it does not support compression from stdin (meaning no "tar | pbzip2"), it does not produce the archives equal to the original bzip2 (although compatible, they are larger), and it felt overall a bit too amateur for me to trust my production backup data to it. So I coded my own one.

This program incorporates the modified libbzip2 sources (part of bzip2). The sources have to be modified because it was not feasible to split the rle compression, block sorting and bit-storing stages apart with the stock library design. This separation was merely hacked in -- to make it the clean way, the library has to be redesigned. This was not the goal, though.

The program was only tested under Linux, kernel 2.6. It should work on any Posix system with pthreads support, but this was not tested, so expect compilation problems. See INSTALL file for details.

The program is meant to be used in production environment. It should be sufficiently stable, but more testing is welcome. I use it myself, but I still don't guarantee you anything. You use it on your own risk, don't blame me if something goes wrong -- send bug reports and patches instead.

bzip2smp 1.0 keywords