bzip2 too slow. Multiple cores are avaible

Question

I am running this command:

pg_dumpall | bzip2 > cluster-$(date --iso).sql.bz2

It takes too long. I look at the processes with top. The bzip2 process takes about 95% and postgres 5% of one core. The wa entry is low. This means the disk is not the bottleneck.

What can I do to increase the performance?

Maybe let bzip2 use more cores. The servers has 16 cores.

Or use an alternative to bzip2?

What can I do to increase the performance?

Unless you need bzip2 for legacy reasons, it's been my personal experience that xz gives better compression/time than bzip2. It's also threaded if you get a new enough program, and it lets you scale time and memory usage from gzipish up to massive depending on what you want. — Perkins, Sep 08 '17 at 20:58
"pigz" is another option - it produces gzip output rather than bzip2 output. And basically everything understands gzip. — Criggie, Sep 08 '17 at 22:31
You might try encrypting it symmetrically with GnuPG with bzip2 compression; it seems surprisingly fast compared with just compressing it, for some unknown reason, even with the highest compression level. It's possible that the algorithm can be faster than the efficiency of my regular compression program, which is GUI-based. — Brōtsyorfuzthrāx, Sep 09 '17 at 09:09
You haven't stated the requirements of your alternative algorithm. Bzip2 is splittable. Is that important to you? — Martin Smith, Sep 09 '17 at 19:53
"*What can I do to increase the performance?*" - not compress it? You don't actually say that you need it compressed, and not-doing-work is always faster than doing-work. Make disk the bottleneck. — TessellatingHeckler, Sep 10 '17 at 02:18
`pigz` has been proposed, and I support the idea; I just wanted to add that the xz version, `pixz`, is about ten times slower, while offering negligible improvement to compression ratio. — Michał Leon, Jun 23 '20 at 10:29

marcelm · Accepted Answer · 2017-09-10T22:31:17.627

There are many compression algorithms around, and bzip2 is one of the slower ones. Plain gzip tends to be significantly faster, at usually not much worse compression. When speed is the most important, lzop is my favourite. Poor compression, but oh so fast.

I decided to have some fun and compare a few algorithms, including their parallel implementations. The input file is the output of pg_dumpall command on my workstation, a 1913 MB SQL file. The hardware is an older quad-core i5. The times are wall-clock times of just the compression. Parallel implementations are set to use all 4 cores. Table sorted by compression speed.

Algorithm     Compressed size        Compression          Decompression

lzop           398MB    20.8%      4.2s    455.6MB/s     3.1s    617.3MB/s
lz4            416MB    21.7%      4.5s    424.2MB/s     1.6s   1181.3MB/s
brotli (q0)    307MB    16.1%      7.3s    262.1MB/s     4.9s    390.5MB/s
brotli (q1)    234MB    12.2%      8.7s    220.0MB/s     4.9s    390.5MB/s
zstd           266MB    13.9%     11.9s    161.1MB/s     3.5s    539.5MB/s
pigz (x4)      232MB    12.1%     13.1s    146.1MB/s     4.2s    455.6MB/s
gzip           232MB    12.1%     39.1s     48.9MB/s     9.2s    208.0MB/s
lbzip2 (x4)    188MB     9.9%     42.0s     45.6MB/s    13.2s    144.9MB/s
pbzip2 (x4)    189MB     9.9%    117.5s     16.3MB/s    20.1s     95.2MB/s
bzip2          189MB     9.9%    273.4s      7.0MB/s    42.8s     44.7MB/s
pixz (x4)      132MB     6.9%    456.3s      4.2MB/s     7.9s    242.2MB/s
xz             132MB     6.9%   1027.8s      1.9MB/s    17.3s    110.6MB/s
brotli (q11)   141MB     7.4%   4979.2s      0.4MB/s     3.6s    531.6MB/s

If the 16 cores of your server are idle enough that all can be used for compression, pbzip2 will probably give you a very significant speed-up. But you need more speed still and you can tolerate ~20% larger files, gzip is probably your best bet.

Update: I added brotli (see TOOGAMs answer) results to the table. brotlis compression quality setting has a very large impact on compression ratio and speed, so I added three settings (q0, q1, and q11). The default is q11, but it is extremely slow, and still worse than xz. q1 looks very good though; the same compression ratio as gzip, but 4-5 times as fast!

Update: Added lbzip2 (see gmathts comment) and zstd (Johnny's comment) to the table, and sorted it by compression speed. lbzip2 puts the bzip2 family back in the running by compressing three times as fast as pbzip2 with a great compression ratio! zstd also looks reasonable but is beat by brotli (q1) in both ratio and speed.

My original conclusion that plain gzip is the best bet is starting to look almost silly. Although for ubiquity, it still can't be beat ;)

For a similar-ish table with far more algorithms, see http://mattmahoney.net/dc/text.html. — Danica, Sep 08 '17 at 23:49
@Dougal Fair enough. My test is on similar data as the OP though (`pg_dumpall` output), so it's probably a bit more representative :) — marcelm, Sep 09 '17 at 00:06
zstd is another one that's missing from the table -- for compressing our log files, I found that a single core zstd process outperforms 16 cores of pbzip2 with comparable compression ratios. — Johnny, Sep 10 '17 at 06:20
`pigz` is a parallelized gzip that works much faster but has comparable compression ratios. — anna328p, Sep 10 '17 at 06:33
@DmitryKudriavtsev Look closely, `pigz` has been in my answer since its very first revision ;) — marcelm, Sep 10 '17 at 12:58
`lz4` is slighty faster and more efficient than `lzop`, by the way. It uses more RAM though, which is relevant in embedded systems. — Daniel B, Sep 10 '17 at 19:06
@DanielB I added `lz4` to the table. In my test, it is slightly slower than `lzop`, but whoo-boy look at the decompression speed :D — marcelm, Sep 10 '17 at 22:31
If you are willing to test multi-threaded versions, you could try `zstd -T4` too. For very fast settings, you can try `zstd -T4 -1`, as `zstd` defaults to `-3`, which is probably the setting you tested. — Cyan, Sep 11 '17 at 18:21
I also found pzstd -1 (parallel zstd) on a multicore machine to be almost as fast as lz4 with better compression ratio — borowis, Sep 07 '18 at 21:09

score 42 · Answer 2 · edited Sep 09 '17 at 05:06

42

Use pbzip2.

The manual says:

pbzip2 is a parallel implementation of the bzip2 block-sorting file compressor that uses pthreads and achieves near-linear speedup on SMP machines. The output of this version is fully compatible with bzip2 v1.0.2 or newer (ie: anything compressed with pbzip2 can be decompressed with bzip2).

It auto-detects the number of processors you have and creates threads accordingly.

edited Sep 09 '17 at 05:06

Cody Gray

135
8

answered Sep 08 '17 at 19:22

ThoriumBR

5,272
2
23
34

That is fine if you are compressing a file, this works horribly through a pipe though – camelccc Sep 08 '17 at 21:38
@camelccc Why do you say that? I don't find that to be the case, at all. You need a fast producer or a large buffer on the pipe in front of it for optimal performance, but that's equally true of `pixz` and `pigz` on a pipe as well. – Michael - sqlbot Sep 08 '17 at 22:22
Depends how large what he is compressing is. If you have a large buffer it's fine as you say, if you are piping something that's much larger than physical ram, I have found things can get rather more interesting. As you say probably true for any compression algorithm though. – camelccc Sep 09 '17 at 00:02
5

bzip2 can use a fair bit of ram, so running 16 bzip workers at a time could consume non-trivial ram, over 1GB. BTW, `lbzip2` seems to give better speed, memory usage and marginally better compression than `pbzip2`. There are benchmarks here: http://vbtechsupport.com/1614/ – gmatht Sep 09 '17 at 04:51
@gmatht `lbzip2` looks nice! I added it to my answer :) – marcelm Sep 10 '17 at 12:58

TOOGAM · Answer 3 · 2017-09-09T16:09:54.357

Google's brotli is a newer format that has gained some wide support within browsers recently, as it has some impressive compression, some impressive speed, and perhaps the most impressive combination/balance of both of those features.
Some data:

Comparison of Brotli, Deflate, Zopfli, LZMA, LZHAM and Bzip2 Compression Algorithms
- e.g., this chart reporting numbers showing Brotli to be roughly 6-14 faster than Bzip2.
CanIUse.com: feature: brotli shows support by Microsoft Edge, Mozilla Firefox, Google Chrome, Apple Safari, Opera (but not Opera Mini or Microsoft Internet Explorer).

Comparison: Brotli vs deflate vs zopfli vs lzma vs lzham vs bzip2
- If you're looking for compression speed, then what you're looking for is which lines are further right on this chart. (The entries to the top of this chart show tight compression ratio. Higher=tighter. However, if compression speed is your priority, then you'll want to pay more attention to what lines reach further right on the chart.)
Comparison: Compression Ratio vs Compression Speed for 7-Zip ZStandard Methods
Facebook's ZStandard is another option, striving to reduce bits but also having a high focus on storing data in a way that reduces missed predictions, thereby allowing for faster speed. Its home page is at: Smaller and faster data compression with ZStandard
Lizard doesn't get quite as high of compression as Brotli or ZStandard, but may be somewhat close in compression ratio, and be quite a bit faster (at least according to this chart which is about speed, although that is reporting decompression)

You didn't mention an operating system. If Windows, 7-Zip with ZStandard (Releases) is a version of 7-Zip that has been modified to provide support for using all of these algorithms.

Interesting, I had heard of `brotli` before, but I forgot about it. I added it to the table of benchmarks in my answer! I was actually a little disappointed with its performance, except at quality setting 1, where it provided the same compression ratio as `gzip` at a much higher speed. — marcelm, Sep 09 '17 at 15:59

score 2 · Answer 4 · answered Sep 10 '17 at 15:37

Use zstd. If it's good enough for Facebook, it's probably good enough for you as well.

On a more serious note, it's actually pretty good. I use it for everything now because it just works, and it lets you trade speed for ratio on a large scale (most often, speed matters more than size anyway since storage is cheap, but speed is a bottleneck).
At compression levels that achieve comparable overall compression as bzip2, it's significantly faster, and if you are willing to pay some extra in CPU time, you can almost achieve results similar to LZMA (although then it will be slower than bzip2). At sligthly worse compression ratios, it is much, much faster than bzip2 or any other mainstream alternative.

Now, your are compressing a SQL dump, which is just about as embarrassingly trivial to compress as it can be. Even the poorest compressors score well on that kind of data.
So you can run zstd with a lower compression level, which will run dozens of times faster and still achieve 95-99% the same compression on that data.

As a bonus, if you will be doing this often and want to invest some extra time, you can "train" the zstd compressor ahead of time, which increases both compression ratio and speed. Note that for training to work well, you will need to feed it individual records, not the whole thing. The way the tool works, it expects many small and somewhat similar samples for training, not one huge blob.

better still, use pzstd (parallel version) on multicore machines — borowis, Sep 07 '18 at 21:11

score 1 · Answer 5 · answered Jul 25 '18 at 11:12

It looks like adjusting (lowering) the block size can have a significant impact on the compression time.

Here are some results of the experiment I did on my machine. I used the time command to measure the execution time. input.txt is a ~250mb text file containing arbitrary json records.

Using the default (biggest) block size (--best merely selects the default behaviour):

# time cat input.txt | bzip2 --best > input-compressed-best.txt.bz

real    0m48.918s
user    0m48.397s
sys     0m0.767s

Using the smallest block size (--fast argument):

# time cat input.txt | bzip2 --fast > input-compressed-fast.txt.bz

real    0m33.859s
user    0m33.571s
sys     0m0.741s

This was a bit surprising discovery, considering that the documntation says:

Compression and decompression speed are virtually unaffected by block size

My current favorite is pbzip2. Have you tried this, too? This question is about an environment where 16 cores are available. — guettli, Jul 25 '18 at 11:37
@guettli unfortunately I have to stick with bzip. I'm using it for Hadoop jobs and bzip is one of the built-in compression there. So in a way it's already parallelised. — Jakub Kukul, Jul 26 '18 at 12:30

bzip2 too slow. Multiple cores are avaible

5 Answers5