S2 Compression

S2 is an extension of Snappy.

S2 is aimed for high throughput, which is why it features concurrent compression for bigger payloads.

Decoding is compatible with Snappy compressed content, but content compressed with S2 cannot be decompressed by Snappy. This means that S2 can seamlessly replace Snappy without converting compressed content.

S2 is designed to have high throughput on content that cannot be compressed. This is important so you don't have to worry about spending CPU cycles on already compressed data.

Benefits over Snappy

• Better compression
• Concurrent stream compression
• Faster decompression
• Ability to quickly skip forward in compressed stream
• Compatible with Snappy compressed content
• Offers alternative, more efficient, but slightly slower compression mode.
• Smaller block size overhead on incompressible blocks.
• Block concatenation
• Automatic stream size padding.

Drawbacks over Snappy

• Not optimized for 32 bit systems.
• No AMD64 assembler implementation yet, meaning slightly slower compression speed on 1 core CPU.
• Uses slightly more memory due to larger blocks and concurrency (configurable).

Usage

Installation: go get -u github.com/klauspost/compress/s2

Full package documentation:

Usage is similar to Snappy.

func EncodeStream(src io.Reader, dst io.Writer) error        
    enc := s2.NewWriter(dst)
    _, err := io.Copy(enc, src)
    if err != nil {
        enc.Close()
        return err
    }
    return enc.Close() 
}

You should always call enc.Close(), otherwise you will leak resources and your encode will be incomplete, as with Snappy.

For the best throughput, you should attempt to reuse the Writer using the Reset() method.

The Writer in S2 is always buffered, therefore NewBufferedWriter in Snappy can be replaced with NewWriter in S2.

func DecodeStream(src io.Reader, dst io.Writer) error        
    dec := s2.NewReader(src)
    _, err := io.Copy(dst, dec)
    return err
}

Similar to the Writer, a Reader can be reused using the Reset method.

For smaller data blocks, there is also a non-streaming interface: Encode(), EncodeBetter() and Decode(). Do however note that these functions (similar to Snappy) does not provide validation of data, so data corruption may be undetected. Stream encoding provides CRC checks of data.

Commandline tools

Some very simply commandline tools are provided; s2c for compression and s2d for decompression.

Binaries can be downloaded on the Releases Page.

Installing then requires Go to be installed. To install them, use:

go install github.com/klauspost/compress/s2/cmd/s2c && go install github.com/klauspost/compress/s2/cmd/s2d

To build binaries to the current folder use:

go build github.com/klauspost/compress/s2/cmd/s2c && go build github.com/klauspost/compress/s2/cmd/s2d

s2c

Usage: s2c [options] file1 file2

Compresses all files supplied as input separately.
Output files are written as 'filename.ext.s2'.
By default output files will be overwritten.
Use - as the only file name to read from stdin and write to stdout.

Wildcards are accepted: testdir/*.txt will compress all files in testdir ending with .txt
Directories can be wildcards as well. testdir/*/*.txt will match testdir/subdir/b.txt

Options:
  -bench int
    	Run benchmark n times. No output will be written
  -blocksize string
    	Max  block size. Examples: 64K, 256K, 1M, 4M. Must be power of two and <= 4MB (default "4M")
  -c	Write all output to stdout. Multiple input files will be concatenated
  -cpu int
    	Compress using this amount of threads (default CPU_THREADS])
  -faster
    	Compress faster, but with a minor compression loss
  -help
    	Display help
  -pad string
    	Pad size to a multiple of this value, Examples: 500, 64K, 256K, 1M, 4M, etc (default "1")
  -q	Don't write any output to terminal, except errors
  -rm
    	Delete source file(s) after successful compression
  -safe
    	Do not overwrite output files

s2d

Usage: s2d [options] file1 file2

Decompresses all files supplied as input. Input files must end with '.s2' or '.snappy'.
Output file names have the extension removed. By default output files will be overwritten.
Use - as the only file name to read from stdin and write to stdout.

Wildcards are accepted: testdir/*.txt will compress all files in testdir ending with .txt
Directories can be wildcards as well. testdir/*/*.txt will match testdir/subdir/b.txt

Options:
  -bench int
    	Run benchmark n times. No output will be written
  -c	Write all output to stdout. Multiple input files will be concatenated
  -help
    	Display help
  -q	Don't write any output to terminal, except errors
  -rm
    	Delete source file(s) after successful decompression
  -safe
    	Do not overwrite output files

Performance

This section will focus on comparisons to Snappy. This package is solely aimed at replacing Snappy as a high speed compression package. If you are mainly looking for better compression zstandard gives better compression, but typically at speeds slightly below "better" mode in this package.

Compression is increased compared to Snappy, mostly around 5-20% and the throughput is typically 25-40% increased (single threaded) compared to the non-assembly Go implementation.

A "better" compression mode is also available. This allows to trade a bit of speed for a minor compression gain. The content compressed in this mode is fully compatible with the standard decoder.

Snappy vs S2 compression speed on 6 core (12 thread) computer, using all threads and a single thread:

File	S2 speed	S2 throughput	S2 % smaller	S2 decomp	S2 "better"	"better" throughput	"better" % smaller	"better" decomp
rawstudio-mint14.tar	7.41x	3401 MB/s	6.98%	1.19x	3.73x	1713 MB/s	10.97%	0.96x
(1 CPU)	1.37x	631 MB/s	-	2085 MB/s	0.67x	309 MB/s	-	1691 MB/s
github-june-2days-2019.json	8.76x	4351 MB/s	28.79%	1.29x	6.64x	3301 MB/s	32.43%	1.23x
(1 CPU)	1.62x	806 MB/s	-	2262 MB/s	1.08x	535 MB/s	-	2153 MB/s
github-ranks-backup.bin	7.70x	3610 MB/s	-5.90%	1.07x	4.65x	2179 MB/s	5.45%	0.93x
(1 CPU)	1.26x	592 MB/s	-	2053 MB/s	0.76x	356 MB/s	-	1796 MB/s
consensus.db.10gb	7.17x	3494 MB/s	14.83%	1.03x	3.43x	1674 MB/s	14.79%	1.03x
(1 CPU)	1.41x	687 MB/s	-	2805 MB/s	0.63x	309 MB/s	-	2796 MB/s
adresser.json	5.16x	4923 MB/s	43.52%	1.17x	4.67x	4456 MB/s	47.15%	1.19x
(1 CPU)	1.76x	1675 MB/s	-	3985 MB/s	1.49x	1425 MB/s	-	4034 MB/s
gob-stream	8.84x	4402 MB/s	22.24%	1.16x	6.58x	3278 MB/s	25.91%	1.08x
(1 CPU)	1.50x	747 MB/s	-	2175 MB/s	1.06x	530 MB/s	-	2039 MB/s
10gb.tar	6.73x	2715 MB/s	1.99%	1.04x	4.50x	1818 MB/s	5.68%	0.91x
(1 CPU)	1.17x	472 MB/s	-	1493 MB/s	0.79x	320 MB/s	-	1312 MB/s
sharnd.out.2gb	0.94x	5987 MB/s	0.01%	0.89x	0.90x	5768 MB/s	0.01%	0.90x
(1 CPU)	1.30x	8323 MB/s	-	4222 MB/s	1.18x	7528 MB/s	-	4266 MB/s
enwik9	10.02x	2337 MB/s	3.66%	1.35x	5.83x	1360 MB/s	15.37%	1.05x
(1 CPU)	1.38x	321 MB/s	-	1230 MB/s	0.84x	197 MB/s	-	956 MB/s
silesia.tar	7.72x	2464 MB/s	5.63%	1.22x	4.65x	1486 MB/s	12.42%	1.01x
(1 CPU)	1.31x	418 MB/s	-	1454 MB/s	0.77x	246 MB/s	-	1210 MB/s

Legend

• S2 speed: Speed of S2 compared to Snappy, using 6 cores and 1 core.
• S2 throughput: Throughput of S2 in MB/s.
• S2 % smaller: How many percent of the Snappy output size is S2 better.
• S2 decomp: Decompression speed of S2 compared to Snappy and absolute speed.
• S2 "better": Speed when enabling "better" compression mode in S2 compared to Snappy.
• "better" throughput: Speed when enabling "better" compression mode in S2 compared to Snappy.
• "better" % smaller: How many percent of the Snappy output size is S2 better when using "better" compression.
• "better" decomp: Decompression speed of S2 "better" mode compared to Snappy and absolute speed.

There is a good speedup across the board when using a single thread and a significant speedup when using multiple threads.

Machine generated data gets by far the biggest compression boost, with size being being reduced by up to 45% of Snappy size.

The "better" compression mode sees a good improvement in all cases, but usually at a performance cost.

We only compare to the non-assembly AMD64 version of Snappy, since S2 does not have an assembly implementation yet. While this may seem to favour S2 on this platform, it is reasonable to assume that an S2 assembly implementation will gain the same speed. Therefore comparing to the non-assembly version gives the best apples-to-apples comparison.

Concurrent Stream Compression

Streams are concurrently compressed. The stream will be distributed among all available CPU cores for the best possible throughput.

Snappy vs S2 compression speed on 6 core (12 thread) computer:

File	S2 throughput	S2 % Smaller	S2 throughput
consensus.db.10gb	7.33x	14.70%	3595 MB/s
github-ranks-backup.bin	7.70x	-5.90%	3610 MB/s
github-june-2days-2019.json	8.76x	28.79%	4351 MB/s
rawstudio-mint14.tar	7.35x	6.98%	3401 MB/s
10gb.tar	6.99x	1.99%	2819 MB/s
enwik9	10.02x	3.66%	2337 MB/s
sharnd.out.2gb	0.94x	0.01%	5987 MB/s
adresser.json	5.16x	45.94%	4923 MB/s
silesia.tar	7.72x	5.63%	2464 MB/s

Incompressible content (sharnd.out.2gb, 2GB random data) sees the smallest speedup. This is likely dominated by synchronization overhead, which is confirmed by the fact that single threaded performance is higher (see above).

Decompression

While the decompression code hasn't changed, there is a significant speedup in decompression speed. S2 prefers longer matches and will typically only find matches that are 6 bytes or longer. While this reduces compression a bit, it improves decompression speed.

The "better" compression mode will actively look for shorter matches, which is why it has a decompression speed quite similar to Snappy.

Single goroutine decompression speed. No assembly:

File	S2 Throughput	S2 throughput
consensus.db.10gb.s2	1.84x	2289.8 MB/s
10gb.tar.s2	1.30x	867.07 MB/s
rawstudio-mint14.tar.s2	1.66x	1329.65 MB/s
github-june-2days-2019.json.s2	2.36x	1831.59 MB/s
github-ranks-backup.bin.s2	1.73x	1390.7 MB/s
enwik9.s2	1.67x	681.53 MB/s
adresser.json.s2	3.41x	4230.53 MB/s
silesia.tar.s2	1.52x	811.58

Single goroutine decompression speed. With AMD64 assembly:

File	S2 throughput	S2 throughput
consensus.db.10gb.s2	1.15x	3074 MB/s
10gb.tar.s2	1.08x	1534 MB/s
rawstudio-mint14.tar.s2	1.27x	2220 MB/s
github-june-2days-2019.json.s2	1.40x	2468 MB/s
github-ranks-backup.bin.s2	1.11x	2132 MB/s
enwik9.s2	1.42x	1280 MB/s
adresser.json.s2	1.34x	4550 MB/s
silesia.tar.s2	1.22x	1374 MB/s

Even though S2 typically compresses better than Snappy, decompression speed is always better.

Block compression

When compressing blocks no concurrent compression is performed just as Snappy. This is because blocks are for smaller payloads and generally will not benefit from concurrent compression.

Benchmarking single block performance is subject to a lot more variation since it only tests a limited number of file patterns. So individual benchmarks should only be seen as a guideline and the overall picture is more important.

An important change is that incompressible blocks will not be more than at most 10 bytes bigger than the input. In rare, worst case scenario Snappy blocks could be significantly bigger than the input.

Standard compression

Absolute Perf	Snappy size	S2 Size	Snappy Speed	Snappy Asm	S2 Speed	Snappy dec	S2 dec
html	22843	21094	583 MB/s	1150 MB/s	814 MB/s	2870 MB/s	3830 MB/s
urls.10K	335492	286835	336 MB/s	514 MB/s	472 MB/s	1600 MB/s	1960 MB/s
fireworks.jpeg	123034	123100	25900 MB/s	24800 MB/s	21600 MB/s	56900 MB/s	63500 MB/s
paper-100k.pdf	85304	84237	5160 MB/s	12400 MB/s	4820 MB/s	24500 MB/s	22300 MB/s
html_x_4	92234	21101	570 MB/s	1080 MB/s	2340 MB/s	2330 MB/s	2370 MB/s
alice29.txt	88034	85934	218 MB/s	425 MB/s	292 MB/s	965 MB/s	1470 MB/s
asyoulik.txt	77503	79569	208 MB/s	392 MB/s	301 MB/s	922 MB/s	1670 MB/s
lcet10.txt	234661	220408	229 MB/s	427 MB/s	298 MB/s	1030 MB/s	1420 MB/s
plrabn12.txt	319267	318197	199 MB/s	354 MB/s	277 MB/s	873 MB/s	1490 MB/s
geo.protodata	23335	18696	743 MB/s	1560 MB/s	1160 MB/s	3820 MB/s	4990 MB/s
kppkn.gtb	69526	65109	335 MB/s	682 MB/s	413 MB/s	1300 MB/s	1470 MB/s

Relative Perf	Snappy size	Asm speed	S2 size improved	S2 Speed	S2 Speed vs. asm	S2 Dec Speed
html	22.31%	1.97x	7.66%	1.40x	0.71x	1.33x
urls.10K	47.78%	1.53x	14.50%	1.40x	0.92x	1.23x
fireworks.jpeg	99.95%	0.96x	-0.05%	0.83x	0.87x	1.12x
paper-100k.pdf	83.30%	2.40x	1.25%	0.93x	0.39x	0.91x
html_x_4	22.52%	1.89x	77.12%	4.11x	2.17x	1.02x
alice29.txt	57.88%	1.95x	2.39%	1.34x	0.69x	1.52x
asyoulik.txt	61.91%	1.88x	-2.67%	1.45x	0.77x	1.81x
lcet10.txt	54.99%	1.86x	6.07%	1.30x	0.70x	1.38x
plrabn12.txt	66.26%	1.78x	0.34%	1.39x	0.78x	1.71x
geo.protodata	19.68%	2.10x	19.88%	1.56x	0.74x	1.31x
kppkn.gtb	37.72%	2.04x	6.35%	1.23x	0.61x	1.13x

In almost all cases there is a size improvement and decompression speed is better, except 1 case.

Except for the PDF and JPEG (mostly incompressible content), the speed is better than the pure Go Snappy. However, looking at the absolute numbers, these cases are still above 4GB/s.

The worst case compression speed of Snappy (plrabn12.txt and asyoulik.txt) is significantly improved.

Better compression

Absolute Perf	Snappy size	Better Size	Snappy Speed	Snappy Asm	Better Speed	Snappy dec	Better dec
html	22843	19867	583 MB/s	1150 MB/s	560 MB/s	2870 MB/s	3100 MB/s
urls.10K	335492	256907	336 MB/s	514 MB/s	291 MB/s	1600 MB/s	1470 MB/s
fireworks.jpeg	123034	123100	25900 MB/s	24800 MB/s	9760 MB/s	56900 MB/s	63300 MB/s
paper-100k.pdf	85304	82915	5160 MB/s	12400 MB/s	594 MB/s	24500 MB/s	15000 MB/s
html_x_4	92234	19875	570 MB/s	1080 MB/s	1750 MB/s	2330 MB/s	2290 MB/s
alice29.txt	88034	73429	218 MB/s	425 MB/s	203 MB/s	965 MB/s	1290 MB/s
asyoulik.txt	77503	66997	208 MB/s	392 MB/s	186 MB/s	922 MB/s	1120 MB/s
lcet10.txt	234661	191969	229 MB/s	427 MB/s	214 MB/s	1030 MB/s	1230 MB/s
plrabn12.txt	319267	272575	199 MB/s	354 MB/s	179 MB/s	873 MB/s	1000 MB/s
geo.protodata	23335	18303	743 MB/s	1560 MB/s	818 MB/s	3820 MB/s	4510 MB/s
kppkn.gtb	69526	61992	335 MB/s	682 MB/s	310 MB/s	1300 MB/s	1260 MB/s

Relative Perf	Snappy size	Asm speed	Better size	Better Speed	Better vs. asm	Better dec
html	22.31%	1.97x	13.03%	0.96x	0.49x	1.08x
urls.10K	47.78%	1.53x	23.42%	0.87x	0.57x	0.92x
fireworks.jpeg	99.95%	0.96x	-0.05%	0.38x	0.39x	1.11x
paper-100k.pdf	83.30%	2.40x	2.80%	0.12x	0.05x	0.61x
html_x_4	22.52%	1.89x	78.45%	3.07x	1.62x	0.98x
alice29.txt	57.88%	1.95x	16.59%	0.93x	0.48x	1.34x
asyoulik.txt	61.91%	1.88x	13.56%	0.89x	0.47x	1.21x
lcet10.txt	54.99%	1.86x	18.19%	0.93x	0.50x	1.19x
plrabn12.txt	66.26%	1.78x	14.62%	0.90x	0.51x	1.15x
geo.protodata	19.68%	2.10x	21.56%	1.10x	0.52x	1.18x
kppkn.gtb	37.72%	2.04x	10.84%	0.93x	0.45x	0.97x

Except for the mostly incompressible JPEG image compression is better and usually in the double digits in terms of percentage reduction over Snappy.

The PDF sample shows a significant slowdown compared to Snappy, as this mode tries harder to compress the data.

Concatenating blocks and streams.

Concatenating streams will concatenate the output of both without recompressing them. While this is inefficient in terms of compression it might be usable in certain scenarios. The 10 byte 'stream identifier' of the second stream can optionally be stripped, but it is not a requirement.

Blocks can be concatenated using the ConcatBlocks function.

Snappy blocks/streams can safely be concatenated with S2 blocks and streams.

Format Extensions

• Frame Stream identifier changed from sNaPpY to S2sTwO.
• Framed compressed blocks can be up to 4MB (up from 64KB).
• Compressed blocks can have an offset of 0, which indicates to repeat the last seen offset.

Repeat offsets must be encoded as a 2.2.1. Copy with 1-byte offset (01), where the offset is 0.

The length is specified by reading the 3-bit length specified in the tag and decode using this table:

Length	Actual Length
0	4
1	5
2	6
3	7
4	8
5	8 + read 1 byte
6	260 + read 2 bytes
7	65540 + read 3 bytes

This allows any repeat offset + length to be represented by 2 to 5 bytes.

Lengths are stored as little endian values.

The first copy of a block cannot be a repeat offset and the offset is not carried across blocks in streams.

Default streaming block size is 1MB.

LICENSE

This code is based on the Snappy-Go implementation.

Use of this source code is governed by a BSD-style license that can be found in the LICENSE file.