perf(query): Improve Intersection with UID pack #8941

harshil-goel · 2023-08-09T10:15:38Z

This PR improves the performance of IntersectCompressedWithBin. LDBC09 Query went from 38 second -> 22 second. Benchmarks show up to 80% improvement in performance.

mangalaman93

Please add unit tests for these functions too

algo/uidlist.go

mangalaman93 · 2023-08-09T16:05:45Z

goos: linux
goarch: amd64
pkg: github.com/dgraph-io/dgraph/algo
cpu: 11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz
len: 
                                                                                          │  in_updated  │
                                                                                          │    sec/op    │
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=100:overlap=0.01:-8       178.3n ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=100:overlap=0.01:-8        1.212µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=100:overlap=0.01:-8          2.447µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=100:overlap=0.01:-8         6.347µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=100:overlap=0.01:-8        45.10µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=1000:overlap=0.01:-8      799.4n ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=1000:overlap=0.01:-8       3.617µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=1000:overlap=0.01:-8         35.03µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=1000:overlap=0.01:-8        74.86µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=1000:overlap=0.01:-8       518.4µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=10000:overlap=0.01:-8     4.953µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=10000:overlap=0.01:-8      44.17µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=10000:overlap=0.01:-8        360.2µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=10000:overlap=0.01:-8       950.5µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=10000:overlap=0.01:-8      4.919m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=100000:overlap=0.01:-8    71.77µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=100000:overlap=0.01:-8     665.1µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=100000:overlap=0.01:-8       3.605m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=100000:overlap=0.01:-8      9.020m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=1000000:overlap=0.01:-8   1.161m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=1000000:overlap=0.01:-8    6.120m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=1000000:overlap=0.01:-8      31.15m ± ∞ ¹
geomean                                                                                     98.72µ
¹ need >= 6 samples for confidence interval at level 0.95

len: 1, compressed: 28, bytes/int: 28.000000
                                                                                          │    in_old    │
                                                                                          │    sec/op    │
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=100:overlap=0.01:-8       701.9n ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=100:overlap=0.01:-8        10.27µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=100:overlap=0.01:-8          62.22µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=100:overlap=0.01:-8         132.2µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=100:overlap=0.01:-8        125.2µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=1000:overlap=0.01:-8      535.6n ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=1000:overlap=0.01:-8       3.268µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=1000:overlap=0.01:-8         1.227m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=1000:overlap=0.01:-8        1.380m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=1000:overlap=0.01:-8       1.157m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=10000:overlap=0.01:-8     4.040µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=10000:overlap=0.01:-8      43.04µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=10000:overlap=0.01:-8        12.65m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=10000:overlap=0.01:-8       13.99m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=10000:overlap=0.01:-8      12.20m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=100000:overlap=0.01:-8    67.35µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=100000:overlap=0.01:-8     612.8µ ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=100000:overlap=0.01:-8       129.2m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=100000:overlap=0.01:-8      117.3m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=1000000:overlap=0.01:-8   1.061m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=1000000:overlap=0.01:-8    6.936m ± ∞ ¹
ListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=1000000:overlap=0.01:-8       1.076 ± ∞ ¹
geomean                                                                                     464.6µ

Old:

len: 1, compressed: 28, bytes/int: 28.000000
goos: linux
goarch: amd64
pkg: github.com/dgraph-io/dgraph/algo
cpu: 11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz
BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=100:overlap=0.01:-8           1698944               701.9 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=100:overlap=0.01:-8             112074             10272 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=100:overlap=0.01:-8                19785             62216 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=100:overlap=0.01:-8                8658            132173 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=100:overlap=0.01:-8               8427            125179 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=1000:overlap=0.01:-8          2230539               535.6 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=1000:overlap=0.01:-8            352585              3268 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=1000:overlap=0.01:-8                 962           1227124 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=1000:overlap=0.01:-8                879           1380495 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=1000:overlap=0.01:-8               962           1156982 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=10000:overlap=0.01:-8          268028              4040 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=10000:overlap=0.01:-8            27942             43035 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=10000:overlap=0.01:-8                 86          12651328 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=10000:overlap=0.01:-8                81          13986664 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=10000:overlap=0.01:-8               86          12201208 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=100000:overlap=0.01:-8          17497             67347 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=100000:overlap=0.01:-8            2047            612767 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=100000:overlap=0.01:-8                 8         129177514 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=100000:overlap=0.01:-8                9         117270228 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=1000000:overlap=0.01:-8           975           1060993 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=1000000:overlap=0.01:-8            172           6935648 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=1000000:overlap=0.01:-8                1        1075782785 ns/op

New:

goos: linux
goarch: amd64
pkg: github.com/dgraph-io/dgraph/algo
cpu: 11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz
BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=100:overlap=0.01:-8           6835880               178.3 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=100:overlap=0.01:-8             923196              1212 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=100:overlap=0.01:-8               458491              2447 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=100:overlap=0.01:-8              197617              6347 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=100:overlap=0.01:-8              27570             45098 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=1000:overlap=0.01:-8          1444035               799.4 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=1000:overlap=0.01:-8            330864              3617 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=1000:overlap=0.01:-8               36864             35029 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=1000:overlap=0.01:-8              14037             74856 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=1000:overlap=0.01:-8              2360            518358 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=10000:overlap=0.01:-8          238930              4953 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=10000:overlap=0.01:-8            35748             44166 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=10000:overlap=0.01:-8               3290            360233 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=10000:overlap=0.01:-8              1342            950549 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=100:size=10000:overlap=0.01:-8              238           4918652 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=100000:overlap=0.01:-8          15912             71766 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=100000:overlap=0.01:-8            1795            665127 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=100000:overlap=0.01:-8               330           3604814 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=10:size=100000:overlap=0.01:-8              133           9019788 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.01:size=1000000:overlap=0.01:-8          1020           1161000 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=0.1:size=1000000:overlap=0.01:-8            170           6119788 ns/op

BenchmarkListIntersectCompressBin/compressed:IntersectWith:ratio=1:size=1000000:overlap=0.01:-8               39          31154933 ns/op

mangalaman93

Great work @harshil-goel. This should improve our query performance too.

all-seeing-code

Lgtm

joshua-goldstein

Looks good 👍

This PR improves the performance of IntersectCompressedWithBin. LDBC09 Query went from 38 second -> 22 second. Benchmarks show up to 80% improvement in performance.

Harshil Goel added 2 commits August 9, 2023 15:17

Added new benchmark

f00f38e

added some changes

d0510cb

harshil-goel requested review from mangalaman93, meghalims, sanjayk-github-dev, all-seeing-code, billprovince and joshua-goldstein as code owners August 9, 2023 10:15

dgraph-bot added area/testing Testing related issues go Pull requests that update Go code labels Aug 9, 2023

mangalaman93 requested changes Aug 9, 2023

View reviewed changes

algo/uidlist.go Outdated Show resolved Hide resolved

algo/uidlist.go Outdated Show resolved Hide resolved

algo/uidlist.go Outdated Show resolved Hide resolved

Harshil Goel added 2 commits August 9, 2023 17:15

fixed something

43e6b6c

Made some changes

ad7944c

mangalaman93 reviewed Aug 9, 2023

View reviewed changes

algo/uidlist.go Show resolved Hide resolved

all-seeing-code previously approved these changes Aug 9, 2023

View reviewed changes

mangalaman93 requested changes Aug 9, 2023

View reviewed changes

algo/uidlist.go Outdated Show resolved Hide resolved

algo/uidlist.go Outdated Show resolved Hide resolved

fixed comments

68a5e9b

harshil-goel dismissed all-seeing-code’s stale review via 68a5e9b August 9, 2023 16:01

updated

d0981eb

mangalaman93 changed the title ~~Updated Implementation of Intersect Compressed with~~ perf(query): Improve Intersection with UID pack Aug 9, 2023

mangalaman93 approved these changes Aug 9, 2023

View reviewed changes

all-seeing-code approved these changes Aug 9, 2023

View reviewed changes

joshua-goldstein approved these changes Aug 9, 2023

View reviewed changes

mangalaman93 merged commit 0474a00 into main Aug 9, 2023

mangalaman93 deleted the harshil-goel/compressed_with branch August 9, 2023 17:06

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

perf(query): Improve Intersection with UID pack #8941

perf(query): Improve Intersection with UID pack #8941

harshil-goel commented Aug 9, 2023 •

edited by mangalaman93

Loading

mangalaman93 left a comment

mangalaman93 commented Aug 9, 2023

mangalaman93 left a comment

all-seeing-code left a comment

joshua-goldstein left a comment

perf(query): Improve Intersection with UID pack #8941

perf(query): Improve Intersection with UID pack #8941

Conversation

harshil-goel commented Aug 9, 2023 • edited by mangalaman93 Loading

mangalaman93 left a comment

Choose a reason for hiding this comment

mangalaman93 commented Aug 9, 2023

mangalaman93 left a comment

Choose a reason for hiding this comment

all-seeing-code left a comment

Choose a reason for hiding this comment

joshua-goldstein left a comment

Choose a reason for hiding this comment

harshil-goel commented Aug 9, 2023 •

edited by mangalaman93

Loading