TurboPFor: Fastest Integer Compression

TurboPFor: The new synonym for "integer compression"
- 100% C (C++ headers), as simple as memcpy
- 👍 Java Critical Natives/JNI. Access TurboPFor incl. SIMD/AVX2! from Java as fast as calling from C
- ✨ FULL range 8/16/32/64 bits scalar + 16/32/64 bits SIMD functions
- No other "Integer Compression" compress/decompress faster
- ✨ Direct Access, integrated (SIMD/AVX2) FOR/delta/Delta of Delta/Zigzag for sorted/unsorted arrays
- 🆕 16 bits + 64 bits SIMD integrated functions
For/PFor/PForDelta
- Novel TurboPFor (PFor/PForDelta) scheme w./ direct access + SIMD/AVX2. 🆕+RLE
- Outstanding compression/speed. More efficient than ANY other fast "integer compression" scheme.
- Compress 70 times faster and decompress up to 4 times faster than OptPFD
Bit Packing
- Fastest and most efficient "SIMD Bit Packing" 10 Billions integers/sec (40Gb/s!)
- Scalar "Bit Packing" decoding nearly as fast as SIMD-Packing in realistic (No "pure cache") scenarios
- Direct/Random Access : Access any single bit packed entry with zero decompression
Variable byte
- Scalar "Variable Byte" faster than ANY other (incl. SIMD) implementation
Simple family
- Novel "Variable Simple" (incl. RLE) faster and more efficient than simple16, simple-8b
Elias fano
- Fastest "Elias Fano" implementation w/ or w/o SIMD/AVX2

Transform
- Scalar & SIMD Transform: Delta, Zigzag, Zigzag of delta, XOR, Transpose/Shuffle,
- 🆕 lossy floating point compression with TurboPFor or TurboTranspose+lz77

Floating Point Compression
- Delta/Zigzag + improved gorilla style + (Differential) Finite Context Method FCM/DFCM floating point compression
- Using TurboPFor, unsurpassed compression and more than 5 GB/s throughput
- 🆕 Error bound lossy floating point compression
🆕 Time Series Compression
- Fastest Gorilla 16/32/64 bits style compression (:new: zigzag of delta + RLE).
- can compress times series to only 0.01%. Speed > 10 GB/s compression and > 13 GB/s decompress.
Inverted Index ...do less, go fast!
- Direct Access to compressed frequency and position data w/ zero decompression
- Novel "Intersection w/ skip intervals", decompress the minimum necessary blocks (~10-15%)!.
- Novel Implicit skips with zero extra overhead
- Novel Efficient Bidirectional Inverted Index Architecture (forward/backwards traversal) incl. "integer compression".
- more than 2000! queries per second on GOV2 dataset (25 millions documents) on a SINGLE core
- ✨ Revolutionary Parallel Query Processing on Multicores > 7000!!! queries/sec on a simple quad core PC.
  ...forget ~~Map Reduce, Hadoop, multi-node clusters,~~ ...

Integer Compression Benchmark:

🆕 Download IcApp a new benchmark for TurboPFor
for testing allmost all integer and floating point file types.
Practical (No PURE cache) "integer compression" benchmark w/ large arrays.
CPU: Skylake i7-6700 3.4GHz gcc 7.2 single thread

- Synthetic data:

Generate and test (zipfian) skewed distribution (100.000.000 integers, Block size=128/256)
Note: Unlike general purpose compression, a small fixed size (ex. 128 integers) is in general used in "integer compression". Large blocks involved, while processing queries (inverted index, search engines, databases, graphs, in memory computing,...) need to be entirely decoded.
```
 ./icbench -a1.5 -m0 -M255 -n100M ZIPF
```

C Size	ratio%	Bits/Integer	C MB/s	D MB/s	Name
62,939,886	15.7	5.04	1588	9400	TurboPFor256
63,392,759	15.8	5.07	1320	6432	TurboPFor
63,392,801	15.8	5.07	1328	924	TurboPForDA
65,060,504	16.3	5.20	60	2748	FP_SIMDOptPFor
65,359,916	16.3	5.23	32	2436	PC_OptPFD
73,477,088	18.4	5.88	408	2484	PC_Simple16
73,481,096	18.4	5.88	624	8748	FP_SimdFastPFor 64Ki *
76,345,136	19.1	6.11	980	2612	VSimple
91,947,533	23.0	7.36	284	11737	QMX 64k *
93,285,864	23.3	7.46	1568	10232	FP_GroupSimple 64Ki *
95,915,096	24.0	7.67	848	3832	Simple-8b
99,910,930	25.0	7.99	13976	11872	TurboPackV
99,910,930	25.0	7.99	9468	9404	TurboPack
99,910,930	25.0	7.99	8420	8876	TurboFor
100,332,929	25.1	8.03	14320	12124	TurboPack256V
101,015,650	25.3	8.08	9520	9484	TurboVByte
102,074,663	25.5	8.17	5712	7916	MaskedVByte
102,074,663	25.5	8.17	2260	4208	PC_Vbyte
102,083,036	25.5	8.17	5200	4268	FP_VByte
112,500,000	28.1	9.00	1528	12140	VarintG8IU
125,000,000	31.2	10.00	4788	11288	StreamVbyte
400,000,000	100.00	32.00	8960	8948	Copy
			N/A	N/A	EliasFano

(*) codecs inefficient for small block sizes are tested with 64Ki integers/block.

MB/s: 1.000.000 bytes/second. 1000 MB/s = 1 GB/s
#BOLD = pareto frontier.
FP=FastPFor SC:simdcomp PC:Polycom
TurboPForDA,TurboForDA: Direct Access is normally used when accessing few individual values.
Eliasfano can be directly used only for increasing sequences

- Data files:

gov2.sorted from DocId data set Block size=128/Delta coding
```
 ./icbench -fS -r gov2.sorted
```

Size	Ratio %	Bits/Integer	C Time MB/s	D Time MB/s	Function
3,321,663,893	13.9	4.44	1320	6088	TurboPFor
3,339,730,557	14.0	4.47	32	2144	PC.OptPFD
3,350,717,959	14.0	4.48	1536	7128	TurboPFor256
3,501,671,314	14.6	4.68	56	2840	VSimple
3,768,146,467	15.8	5.04	3228	3652	EliasFanoV
3,822,161,885	16.0	5.11	572	2444	PC_Simple16
4,521,326,518	18.9	6.05	836	3296	Simple-8b
4,649,671,427	19.4	6.22	3084	3848	TurboVbyte
4,955,740,045	20.7	6.63	7064	10268	TurboPackV
4,955,740,045	20.7	6.63	5724	8020	TurboPack
5,205,324,760	21.8	6.96	6952	9488	SC_SIMDPack128
5,393,769,503	22.5	7.21	9912	11588	TurboPackV256
6,221,886,390	26.0	8.32	6668	6952	TurboFor
6,221,886,390	26.0	8.32	6644	2260	TurboForDA
6,699,519,000	28.0	8.96	1888	1980	FP_Vbyte
6,700,989,563	28.0	8.96	2740	3384	MaskedVByte
7,622,896,878	31.9	10.20	836	4792	VarintG8IU
8,060,125,035	33.7	11.50	3536	8684	Streamvbyte
8,594,342,216	35.9	11.50	5228	6376	libfor
23,918,861,764	100.0	32.00	5824	5924	Copy

Block size: 64Ki = 256k bytes. Ki=1024 Integers

Size	Ratio %	Bits/Integer	C Time MB/s	D Time MB/s	Function
3,164,940,562	13.2	4.23	1344	6004	TurboPFor 64Ki
3,273,213,464	13.7	4.38	1496	7008	TurboPFor256 64Ki
3,965,982,954	16.6	5.30	1520	2452	lz4+DT 64Ki
4,234,154,427	17.7	5.66	436	5672	qmx 64Ki
6,074,995,117	25.4	8.13	1976	2916	blosc_lz4 64Ki
8,773,150,644	36.7	11.74	2548	5204	blosc_lz 64Ki

"lz4+DT 64Ki" = Delta+Transpose from TurboPFor + lz4
"blosc_lz4" internal lz4 compressor+vectorized shuffle

- Time Series:

Test file Timestamps: ts.txt(sorted)
```
  ./icapp -Ft ts.txt -I15 -J15
```

Function	C MB/s	size	ratio%	D MB/s	Text
bvzenc32	10632	45,909	0.008	12823	ZigZag
bvzzenc32	8914	56,713	0.010	13499	ZigZag Delta of delta
vsenc32	12294	140,400	0.024	12877	Variable Simple
p4nzenc256v32	1932	596,018	0.10	13326	TurboPFor256 ZigZag
p4ndenc256v32	1961	596,018	0.10	13339	TurboPFor256 Delta
bitndpack256v32	12564	909,189	0.16	13505	TurboPackV256 Delta
p4nzenc32	1810	1,159,633	0.20	8502	TurboPFor ZigZag
p4nzenc128v32	1795	1,159,633	0.20	13338	TurboPFor ZigZag
bitnzpack256v32	9651	1,254,757	0.22	13503	TurboPackV256 ZigZag
bitnzpack128v32	10155	1,472,804	0.26	13380	TurboPackV ZigZag
vbddenc32	6198	18,057,296	3.13	10982	TurboVByte Delta of delta
memcpy	13397	577,141,992	100.00

- Transpose/Shuffle (no compression)

    ./icbench -eTRANSFORM ZIPF

Size	C Time MB/s	D Time MB/s	Function
100,000,000	9400	9132	TPbyte 4 TurboPFor Byte Transpose/shuffle AVX2
100,000,000	8784	8860	TPbyte 4 TurboPFor Byte Transpose/shuffle SSE
100,000,000	7688	7656	Blosc_Shuffle AVX2
100,000,000	5204	7460	TPnibble 4 TurboPFor Nibble Transpose/shuffle SSE
100,000,000	6620	6284	Blosc shuffle SSE
100,000,000	3156	3372	Bitshuffle AVX2
100,000,000	2100	2176	Bitshuffle SSE

- (Lossy) Floating point compression:

    ./icapp -Fd file          " 64 bits floating point raw file 
    ./icapp -Ff file          " 32 bits floating point raw file 
    ./icapp -Fcf file         " text file with miltiple entries (ex.  8.657,56.8,4.5 ...)
    ./icapp -Ftf file         " text file (1 entry/line)
    ./icapp -Ftf file -v5     " + display the first entries read
    ./icapp -Ftf file.csv -K3 " but 3th column in a csv file (ex. number,Text,456.5 -> 456.5
    ./icapp -Ftf file -g.001  " lossy compression with allowed error 0.001

see also TurboTranspose

- Compressed Inverted Index Intersections with GOV2

GOV2: 426GB, 25 Millions documents, average doc. size=18k.

Aol query log: 18.000 queries
~1300 queries per second (single core)
~5000 queries per second (quad core)
Ratio = 14.37% Decoded/Total Integers.
TREC Million Query Track (1MQT):
~1100 queries per second (Single core)
~4500 queries per second (Quad core CPU)
Ratio = 11.59% Decoded/Total Integers.

Benchmarking intersections (Single core, AOL query log)

max.docid/q	Time s	q/s	ms/q	% docid found
1.000	7.88	2283.1	0.438	81
10.000	10.54	1708.5	0.585	84
ALL	13.96	1289.0	0.776	100
q/s: queries/second, ms/q:milliseconds/query

Benchmarking Parallel Query Processing (Quad core, AOL query log)

max.docid/q	Time s	q/s	ms/q	% docids found
1.000	2.66	6772.6	0.148	81
10.000	3.39	5307.5	0.188	84
ALL	3.57	5036.5	0.199	100

Notes:

Search engines are spending 90% of the time in intersections when processing queries.
Most search engines are using pruning strategies, caching popular queries,... to reduce the time for intersections and query processing.
As indication, google is processing 40.000 Queries per seconds, using 900.000 multicore servers for searching 8 billions web pages (320 X size of GOV2).
Recent "integer compression" GOV2 experiments (best paper at ECIR 2014) On Inverted Index Compression for Search Engine Efficiency using 8-core Xeon PC are reporting 1.2 seconds per query (for 1.000 Top-k docids).

Compile:

    Download or clone TurboPFor
	git clone git://github.com/powturbo/TurboPFor.git
	cd TurboPFor
    
    To benchmark external libraries:
	git clone --recursive git://github.com/powturbo/TurboPFor.git
	cd TurboPFor

Linux, Windows (MingW), Clang,... (see also makefile)

	make
	or
	make AVX2=1

	Include external libs
	make CODEC1=1 CODEC2=1 

	Disable SIMD
	make NSIMD=1

Windows visual c++

	nmake /f makefile.vs

Testing:

- Synthetic data (use ZIPF parameter):

benchmark groups of "integer compression" functions

./icbench -eBENCH -a1.2 -m0 -M255 -n100M ZIPF
./icbench -eBITPACK/VBYTE -a1.2 -m0 -M255 -n100M ZIPF

Type "icbench -l1" for a list

-zipfian distribution alpha = 1.2 (Ex. -a1.0=uniform -a1.5=skewed distribution)
-number of integers = 100.000.000
-integer range from 0 to 255

Unsorted lists: individual function test (ex. Copy TurboPack TurboPFor)

./icbench -a1.5 -m0 -M255 -ecopy/turbopack/turbopfor/turbopack256v ZIPF

Unsorted lists: Zigzag encoding w/ option -fz or FOR encoding

./icbench -fz -eturbovbyte/turbopfor/turbopackv ZIPF
./icbench -eturboforv ZIPF

Sorted lists: differential coding w/ option -fs (increasing) or -fS (strictly increasing)
```
./icbench -fs -eturbopack/turbopfor/turbopfor256v ZIPF
```
Generate interactive "file.html" plot for browsing
```
./icbench -p2 -S2 -Q3 file.tbb
```

Unit test: test function from bit size 0 to 32

./icbench -m0 -M32 -eturbpfor 
./icbench -m0 -M8 -eturbopack -fs -n1M

- Data files:

Raw 32 bits binary data file Test data

./icbench file
./icapp file           
./icapp -Fs file         "16 bits binary file
./icapp -Fu file         "32 bits binary file
./icapp -Fl file         "64 bits binary file
./icapp -Ff file         "32 bits floating point binary file
./icapp -Fd file         "64 bits floating point binary file

Text file: 1 entry per line. Test data: ts.txt(sorted) and lat.txt(unsorted))

./icbench -eBENCH -fts ts.txt
./icbench -eBENCH -ft  lat.txt

./icapp -Fts data.txt            "text file, one 16 bits integer per line
./icapp -Ftu ts.txt              "text file, one 32 bits integer per line
./icapp -Ftl ts.txt              "text file, one 64 bits integer per line
./icapp -Ftf file                "text file, one 32 bits floating point (ex. 8.32456) per line
./icapp -Ftd file                "text file, one 64 bits floating point (ex. 8.324567789) per line
./icapp -Ftd file -v5            "like prev., display the first 100 values read
./icapp -Ftd file -v5 -g.00001   "like prev., error bound lossy floating point compression
./icapp -Ftt file                "text file, timestamp in seconds iso-8601 -> 32 bits integer (ex. 2018-03-12T04:31:06)
./icapp -FtT file                "text file, timestamp in milliseconds iso-8601 -> 64 bits integer (ex. 2018-03-12T04:31:06.345)
./icapp -Ftl -D2 -H file         "skip 1th line, convert numbers with 2 decimal digits to 64 bits integers (ex. 456.23 -> 45623)
./icapp -Ftl -D2 -H -K3 file.csv  "like prev., use the 3th number in the line (ex. label=3245, text=99 usage=456.23 -> 456.23 )
./icapp -Ftl -D2 -H -K3 -k| file.csv "like prev., use '|' as separator

Text file: multiple numbers separated by non-digits (0..9,-,.) characters (ex. 134534,-45678,98788,4345, )

./icapp -Fc data.txt         "text file, 32 bits integers (ex. 56789,3245,23,678 ) 
./icapp -Fcd data.txt        "text file, 64 bits floting-point numbers (ex. 34.7689,5.20,45.789 )

Multiblocks of 32 bits binary file. (Example gov2 from DocId data set)
Block format: [n1: #of Ids][Id1] [Id2]...[IdN] [n2: #of Ids][Id1][Id2]...[IdN]...
```
./icbench -fS -r gov2.sorted
```

- Intersections:

1 - Download Gov2 (or ClueWeb09) + query files (Ex. "1mq.txt") from DocId data set
8GB RAM required (16GB recommended for benchmarking "clueweb09" files).

2 - Create index file

    ./idxcr gov2.sorted .

create inverted index file "gov2.sorted.i" in the current directory

3 - Test intersections

    ./idxqry gov2.sorted.i 1mq.txt

run queries in file "1mq.txt" over the index of gov2 file

- Parallel Query Processing:

1 - Create partitions

    ./idxseg gov2.sorted . -26m -s8

create 8 (CPU hardware threads) partitions for a total of ~26 millions document ids

2 - Create index file for each partition

  ./idxcr gov2.sorted.s*

create inverted index file for all partitions "gov2.sorted.s00 - gov2.sorted.s07" in the current directory

3 - Intersections:

delete "idxqry.o" file and then type "make para" to compile "idxqry" w. multithreading

  ./idxqry gov2.sorted.s*.i 1mq.txt

run queries in file "1mq.txt" over the index of all gov2 partitions "gov2.sorted.s00.i - gov2.sorted.s07.i".

Function usage:

See benchmark "icbench" program for "integer compression" usage examples. In general encoding/decoding functions are of the form:

char *endptr = encode( unsigned *in, unsigned n, char *out, [unsigned start], [int b])
endptr : set by encode to the next character in "out" after the encoded buffer
in : input integer array
n : number of elements
out : pointer to output buffer
b : number of bits. Only for bit packing functions
start : previous value. Only for integrated delta encoding functions

char *endptr = decode( char *in, unsigned n, unsigned *out, [unsigned start], [int b])
endptr : set by decode to the next character in "in" after the decoded buffer
in : pointer to input buffer
n : number of elements
out : output integer array
b : number of bits. Only for bit unpacking functions
start : previous value. Only for integrated delta decoding functions

Simple high level functions:

size_t compressed_size = encode( unsigned *in, size_t n, char *out)
compressed_size : number of bytes written into compressed output buffer out

size_t compressed_size = decode( char *in, size_t n, unsigned *out)
compressed_size : number of bytes read from compressed input buffer in

Function syntax:

{vb | p4 | bit | vs}[d | d1 | f | fm | z ]{enc/dec | pack/unpack}[| 128V | 256V][8 | 16 | 32 | 64]:
vb: variable byte
p4: turbopfor
vs: variable simple
bit: bit packing

d: delta encoding for increasing integer lists (sorted w/ duplicate)
d1: delta encoding for strictly increasing integer lists (sorted unique)
f : FOR encoding for sorted integer lists
fm: FOR encoding for unsorted integer lists
z: ZigZag encoding for unsorted integer lists

enc/pack: encode
dec/unpack:decode
XX : integer size (8/16/32/64)

header files to use with documentation:

c/c++ header file	Integer Compression functions	examples
vint.h	variable byte	vbenc32/vbdec32 vbdenc32/vbddec32 vbzenc32/vbzdec32
vsimple.h	variable simple	vsenc64/vsdec64
vp4.h	TurboPFor	p4enc32/p4dec32 p4denc32/p4ddec32 p4zenc32/p4zdec32
bitpack.h	Bit Packing, For, +Direct Access	bitpack256v32/bitunpack256v32 bitforenc64/bitfordec64
eliasfano.h	Elias Fano	efanoenc256v32/efanoc256v32

Environment:

OS/Compiler (64 bits):

Linux: GNU GCC (>=4.6)
clang (>=3.2)
Windows: MinGW-w64 (no parallel query processing demo app)
Visual c++ (VS2008-VS2017)

Multithreading:

All TurboPFor integer compression functions are thread safe

References:

Benchmark references:
- FastPFor + Simdcomp: SIMDPack FPF, Vbyte FPF, VarintG8IU, StreamVbyte, GroupSimple
- Optimized Pfor-delta compression code: OptPFD/OptP4, Simple16 (limited to 28 bits integers)
- MaskedVByte. See also: Vectorized VByte Decoding
- Streamvbyte.
- Index Compression Using 64-Bit Words: Simple-8b (speed optimized version tested)
- libfor
- Compression, SIMD, and Postings Lists QMX integer compression from the "simple family"
- lz4. included w. block size 64K as indication. Tested after preprocessing w. delta+transpose
- blosc. blosc is like transpose/shuffle+lz77. Tested blosc+lz4 and blosclz incl. vectorizeed shuffle.
- Document identifier data set
Integer compression publications:
Applications:

Last update: 16 Mar 2018

Name		Name	Last commit message	Last commit date
Latest commit History 1,125 Commits
ext		ext
java		java
vs		vs
.gitattributes		.gitattributes
.gitmodules		.gitmodules
.travis.yml		.travis.yml
README.md		README.md
bitpack.c		bitpack.c
bitpack.h		bitpack.h
bitpack_.h		bitpack_.h
bitunpack.c		bitunpack.c
bitunpack_.h		bitunpack_.h
bitutil.c		bitutil.c
bitutil.h		bitutil.h
conf.h		conf.h
eliasfano.c		eliasfano.c
eliasfano.h		eliasfano.h
fp.c		fp.c
fp.h		fp.h
icapp.c		icapp.c
icbench.c		icbench.c
idx.h		idx.h
idxcr.c		idxcr.c
idxqry.c		idxqry.c
idxseg.c		idxseg.c
jic.c		jic.c
makefile		makefile
makefile.vs		makefile.vs
plugins.cc		plugins.cc
plugins.h		plugins.h
time_.h		time_.h
transpose.c		transpose.c
transpose.h		transpose.h
vint.c		vint.c
vint.h		vint.h
vp4.h		vp4.h
vp4c.c		vp4c.c
vp4d.c		vp4d.c
vsimple.c		vsimple.c
vsimple.h		vsimple.h

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

TurboPFor: Fastest Integer Compression

Integer Compression Benchmark:

- Synthetic data:

- Data files:

- Time Series:

- Transpose/Shuffle (no compression)

- (Lossy) Floating point compression:

- Compressed Inverted Index Intersections with GOV2

Notes:

Compile:

Linux, Windows (MingW), Clang,... (see also makefile)

Windows visual c++

Testing:

- Synthetic data (use ZIPF parameter):

- Data files:

- Intersections:

- Parallel Query Processing:

Function usage:

Function syntax:

Environment:

OS/Compiler (64 bits):

Multithreading:

References:

About

Releases

Packages

Languages

vesslanjin/TurboPFor

Folders and files

Latest commit

History

Repository files navigation

TurboPFor: Fastest Integer Compression

Integer Compression Benchmark:

- Synthetic data:

- Data files:

- Time Series:

- Transpose/Shuffle (no compression)

- (Lossy) Floating point compression:

- Compressed Inverted Index Intersections with GOV2

Notes:

Compile:

Linux, Windows (MingW), Clang,... (see also makefile)

Windows visual c++

Testing:

- Synthetic data (use ZIPF parameter):

- Data files:

- Intersections:

- Parallel Query Processing:

Function usage:

Function syntax:

Environment:

OS/Compiler (64 bits):

Multithreading:

References:

About

Resources

Stars

Watchers

Forks

Releases

Packages 0

Languages

Packages