use Nat64 for PRNG #69

chenyan-dfinity · 2023-07-22T01:12:36Z

Reduce overhead in PRNG

Don't converting Nat32 to Nat to avoid allocation
Remove debug_show
Switch to Nat64 to avoid wrapping
Reuse rand for removal
Scale collection benchmarks to 1m

github-actions · 2023-07-22T01:50:03Z

Note
Diffing the performance result against the published result from main branch.
Unchanged benchmarks are omitted.

Map

	binary_size	generate 50k	max mem	batch_get 50	batch_put 50	batch_remove 50
hashmap	133_828 ($\textcolor{green}{-12.29\%}$)	344_919_719 ($\textcolor{green}{-71.15\%}$)	3_099_888 ($\textcolor{green}{-65.94\%}$)	283_600 ($\textcolor{green}{-48.02\%}$)	273_010_658 ($\textcolor{green}{-25.32\%}$)	304_884 ($\textcolor{green}{-41.47\%}$)
triemap	135_316 ($\textcolor{green}{-13.49\%}$)	476_664_789 ($\textcolor{green}{-64.40\%}$)	3_713_736 ($\textcolor{green}{-61.78\%}$)	187_068 ($\textcolor{green}{-59.31\%}$)	447_552 ($\textcolor{green}{-62.49\%}$)	431_270 ($\textcolor{green}{-37.18\%}$)
rbtree	136_114 ($\textcolor{green}{-11.19\%}$)	247_816_636 ($\textcolor{green}{-77.78\%}$)	2_899_996 ($\textcolor{green}{-67.42\%}$)	72_364 ($\textcolor{green}{-79.60\%}$)	214_369 ($\textcolor{green}{-77.77\%}$)	227_128 ($\textcolor{green}{-54.13\%}$)
splay	131_868 ($\textcolor{green}{-13.64\%}$)	454_525_330 ($\textcolor{green}{-65.66\%}$)	2_699_932 ($\textcolor{green}{-68.97\%}$)	437_199 ($\textcolor{green}{-39.20\%}$)	463_233 ($\textcolor{green}{-61.85\%}$)	639_212 ($\textcolor{green}{-10.87\%}$)
btree	176_459 ($\textcolor{green}{-2.09\%}$)	351_887_997 ($\textcolor{green}{-71.22\%}$)	1_554_008 ($\textcolor{green}{-79.43\%}$)	219_323 ($\textcolor{green}{-56.39\%}$)	337_458 ($\textcolor{green}{-69.05\%}$)	368_138 ($\textcolor{green}{-31.88\%}$)
zhenya_hashmap	141_855 ($\textcolor{green}{-4.46\%}$)	131_227_951 ($\textcolor{green}{-86.74\%}$)	3_299_636 ($\textcolor{green}{-64.53\%}$)	62_085 ($\textcolor{green}{-81.46\%}$)	76_830 ($\textcolor{green}{-90.61\%}$)	89_171 ($\textcolor{green}{-73.40\%}$)
btreemap_rs	413_478 ($\textcolor{green}{-4.63\%}$)	70_011_959 ($\textcolor{green}{-37.04\%}$)	1_245_184 ($\textcolor{green}{-24.00\%}$)	57_133 ($\textcolor{green}{-1.18\%}$)	86_370 ($\textcolor{green}{-34.05\%}$)	79_811 ($\textcolor{red}{31.03\%}$)
hashmap_rs	406_096 ($\textcolor{green}{-4.55\%}$)	15_433_930 ($\textcolor{green}{-67.51\%}$)	1_835_008	17_154 ($\textcolor{green}{-2.99\%}$)	21_926 ($\textcolor{green}{-59.99\%}$)	19_973 ($\textcolor{red}{9.16\%}$)

Priority queue

	binary_size	heapify 50k	max mem	pop_min 50	put 50
heap	129_512 ($\textcolor{green}{-7.46\%}$)	84_138_307 ($\textcolor{green}{-77.23\%}$)	1_499_928 ($\textcolor{red}{7.14\%}$)	336_956 ($\textcolor{red}{0.77\%}$)	115_668 ($\textcolor{green}{-70.90\%}$)
heap_rs	403_925 ($\textcolor{green}{-0.09\%}$)	6_215_095 ($\textcolor{red}{19.00\%}$)	1_114_112 ($\textcolor{red}{36.00\%}$)	45_601 ($\textcolor{green}{-0.60\%}$)	18_370 ($\textcolor{green}{-1.46\%}$)

MoVM

	binary_size	generate 10k	max mem	batch_get 50	batch_put 50	batch_remove 50
hashmap	133_828 ($\textcolor{green}{-12.29\%}$)	68_512_644 ($\textcolor{green}{-71.33\%}$)	620_540 ($\textcolor{green}{-65.92\%}$)	281_892 ($\textcolor{green}{-48.18\%}$)	54_360_484 ($\textcolor{green}{-26.07\%}$)	303_285 ($\textcolor{green}{-41.52\%}$)
hashmap_rs	406_096 ($\textcolor{green}{-4.55\%}$)	3_429_927 ($\textcolor{green}{-65.00\%}$)	950_272	16_522 ($\textcolor{green}{-2.99\%}$)	20_832 ($\textcolor{green}{-61.52\%}$)	19_973 ($\textcolor{red}{15.86\%}$)
imrc_hashmap_rs	413_588 ($\textcolor{green}{-4.68\%}$)	18_594_807 ($\textcolor{green}{-27.29\%}$)	1_507_328 ($\textcolor{green}{-4.17\%}$)	29_484 ($\textcolor{red}{3.70\%}$)	107_838 ($\textcolor{green}{-27.81\%}$)	80_154 ($\textcolor{red}{120.91\%}$)
movm_rs	1_712_108 ($\textcolor{green}{-0.01\%}$)	1_089_653_746	2_654_208	2_497_600	6_977_084	5_505_446
movm_dynamic_rs	1_830_967 ($\textcolor{green}{-0.00\%}$)	509_925_551	2_129_920	2_045_671	2_754_013	2_043_229

Statistics

binary_size: -6.36% [-8.65%, -4.08%]
max_mem: -41.73% [-62.95%, -20.51%]
cycles: -39.22% [-48.56%, -29.88%]

Basic DAO

	binary_size	init	transfer_token	submit_proposal	vote_proposal
Motoko	225_805	37_469 ($\textcolor{green}{-0.06\%}$)	16_228 ($\textcolor{green}{-0.26\%}$)	12_658 ($\textcolor{red}{0.02\%}$)	14_128 ($\textcolor{red}{0.16\%}$)
Rust	704_886	471_865	86_470	104_617	115_765

DIP721 NFT

Note
Same as main branch, skipping.

Statistics

binary_size: no change
max_mem: no change
cycles: -0.04% [-0.24%, 0.17%]

Heartbeat

	binary_size	heartbeat
Motoko	118_909	7_392
Rust	23_699	474 ($\textcolor{green}{-40.08\%}$)

Timer

Note
Same as main branch, skipping.

Statistics

binary_size: no change
max_mem: no change
cycles: no change

Garbage Collection

	generate 80k	max mem	batch_get 50	batch_put 50	batch_remove 50
default	101_238_593 ($\textcolor{green}{-59.81\%}$)	5_940_184 ($\textcolor{green}{-61.77\%}$)	50	50	50
copying	101_238_543 ($\textcolor{green}{-59.81\%}$)	5_940_184 ($\textcolor{green}{-61.77\%}$)	101_236_505 ($\textcolor{green}{-59.81\%}$)	101_303_512 ($\textcolor{green}{-59.81\%}$)	101_238_661 ($\textcolor{green}{-59.84\%}$)
compacting	165_791_998 ($\textcolor{green}{-56.98\%}$)	5_940_184 ($\textcolor{green}{-61.77\%}$)	129_325_699 ($\textcolor{green}{-56.28\%}$)	153_305_309 ($\textcolor{green}{-56.78\%}$)	155_891_703 ($\textcolor{green}{-54.03\%}$)
generational	248_396_502 ($\textcolor{green}{-57.91\%}$)	5_940_448 ($\textcolor{green}{-61.77\%}$)	47_649 ($\textcolor{green}{-6.94\%}$)	850_111 ($\textcolor{green}{-19.14\%}$)	762_565 ($\textcolor{red}{28.28\%}$)
incremental	253_634_373 ($\textcolor{red}{31.65\%}$)	4_624 ($\textcolor{green}{-0.09\%}$)	17_416_620 ($\textcolor{green}{-96.65\%}$)	290_834_822 ($\textcolor{red}{124.03\%}$)	27_313_007 ($\textcolor{green}{-91.52\%}$)

Actor class

Note
Same as main branch, skipping.

Statistics

binary_size: no change
max_mem: -49.44% [-75.74%, -23.13%]
cycles: -35.96% [-58.94%, -12.98%]

Overall Statistics

binary_size: -6.36% [-8.65%, -4.08%]
max_mem: -44.14% [-59.45%, -28.82%]
cycles: -36.23% [-44.72%, -27.75%]

github-actions · 2023-07-22T01:50:04Z

Note
The flamegraph link only works after you merge.
Unchanged benchmarks are omitted.

Collection libraries

Measure different collection libraries written in both Motoko and Rust.
The library names with _rs suffix are written in Rust; the rest are written in Motoko.

We use the same random number generator with fixed seed to ensure that all collections contain
the same elements, and the queries are exactly the same. Below we explain the measurements of each column in the table:

generate 1m. Insert 1m Nat64 integers into the collection. For Motoko collections, it usually triggers the GC; the rest of the column are not likely to trigger GC.
max mem. For Motoko, it reports rts_max_live_size after generate call; For Rust, it reports the Wasm's memory page * 32Kb.
batch_get 50. Find 50 elements from the collection.
batch_put 50. Insert 50 elements to the collection.
batch_remove 50. Remove 50 elements from the collection.

💎 Takeaways

The platform only charges for instruction count. Data structures which make use of caching and locality have no impact on the cost.
We have a limit on the maximal cycles per round. This means asymptotic behavior doesn't matter much. We care more about the performance up to a fixed N. In the extreme cases, you may see an O(10000 nlogn) algorithm hitting the limit, while an O(n^2) algorithm runs just fine.
Amortized algorithms/GC may need to be more eager to avoid hitting the cycle limit on a particular round.
Rust costs more cycles to process complicated Candid data, but it is more efficient in performing core computations.

Note

The Candid interface of the benchmark is minimal, therefore the serialization cost is negligible in this measurement.

Due to the instrumentation overhead and cycle limit, we cannot profile computations with large collections. Hopefully, when deterministic time slicing is ready, we can measure the performance on larger memory footprint.

hashmap uses amortized data structure. When the initial capacity is reached, it has to copy the whole array, thus the cost of batch_put 50 is much higher than other data structures.

btree comes from Byron Becker's stable BTreeMap library.

zhenya_hashmap comes from Zhenya Usenko's stable HashMap library.

hashmap_rs uses the fxhash crate, which is the same as std::collections::HashMap, but with a deterministic hasher. This ensures reproducible result.

imrc_hashmap_rs uses the im-rc crate, which is the immutable version hashmap in Rust.

The MoVM table measures the performance of an experimental implementation of Motoko interpreter. External developers can ignore this table for now.

Map

	binary_size	generate 1m	max mem	batch_get 50	batch_put 50	batch_remove 50
hashmap	133_828	6_960_077_358	61_987_708	287_469	5_515_887_135	308_972
triemap	135_316	11_431_084_368	74_215_992	222_768	547_650	538_998
rbtree	136_114	5_979_229_531	57_995_880	88_900	268_568	278_334
splay	131_868	11_568_250_397	53_995_816	551_921	581_659	810_215
btree	176_459	8_224_241_532	31_103_868	277_537	384_166	429_036
zhenya_hashmap	141_855	2_756_209_728	65_987_456	68_392	83_100	93_270
btreemap_rs	413_478	1_649_709_879	13_762_560	66_814	112_263	81_263
imrc_hashmap_rs	413_588	2_385_702_121	122_454_016	32_846	162_715	98_494
hashmap_rs	406_096	392_593_368	36_536_320	16_498	20_863	19_973

Priority queue

	binary_size	heapify 1m	max mem	pop_min 50	put 50
heap	127_748	4_684_517_789	29_995_812	511_494	186_460
heap_rs	403_925	123_102_482	9_109_504	53_320	18_138

Sample Dapps

Measure the performance of some typical dapps:

Basic DAO,
with heartbeat disabled to make profiling easier. We have a separate benchmark to measure heartbeat performance.
DIP721 NFT

Note

The cost difference is mainly due to the Candid serialization cost.

Motoko statically compiles/specializes the serialization code for each method, whereas in Rust, we use serde to dynamically deserialize data based on data on the wire.

We could improve the performance on the Rust side by using parser combinators. But it is a challenge to maintain the ergonomics provided by serde.

For real-world applications, we tend to send small data for each endpoint, which makes the Candid overhead in Rust tolerable.

Basic DAO

	binary_size	init	transfer_token	submit_proposal	vote_proposal
Motoko	225_805	37_517	16_312	12_696	14_153
Rust	704_886	471_865	86_525	104_617	115_765

DIP721 NFT

	binary_size	init	mint_token	transfer_token
Motoko	183_882	12_181	22_319	4_710
Rust	766_710	125_034	324_482	77_116

Heartbeat / Timer

Measure the cost of empty heartbeat and timer job.

setTimer measures both the setTimer(0) method and the execution of empty job.
It is not easy to reliably capture the above events in one flamegraph, as the implementation detail
of the replica can affect how we measure this. Typically, a correct flamegraph contains both setTimer and canister_global_timer function. If it's not there, we may need to adjust the script.

Heartbeat

	binary_size	heartbeat
Motoko	118_909	7_392
Rust	23_699	474

Timer

	binary_size	setTimer	cancelTimer
Motoko	125_168	15_208	1_679
Rust	434_848	43_540	7_683

Motoko Specific Benchmarks

Measure various features only available in Motoko.

Garbage Collection. Measure Motoko garbage collection cost using the Triemap benchmark. The max mem column reports rts_max_live_size after generate call. The cycle cost numbers reported here are garbage collection cost only. Some flamegraphs are truncated due to the 2M log size limit. The dfx/ic-wasm optimizer is disabled for the garbage collection test cases due to how the optimizer affects function names, making profiling trickier.
- default. Compile with the default GC option. With the current GC scheduler, generate will trigger the copying GC. The rest of the methods will not trigger GC.
- copying. Compile with --force-gc --copying-gc.
- compacting. Compile with --force-gc --compacting-gc.
- generational. Compile with --force-gc --generational-gc.
- incremental. Compile with --force-gc --incremental-gc.
Actor class. Measure the cost of spawning actor class, using the Actor classes example.

Garbage Collection

	generate 800k	max mem	batch_get 50	batch_put 50	batch_remove 50
default	1_012_258_524	59_396_752	50	50	50
copying	1_012_258_474	59_396_752	1_012_236_033	1_012_303_043	1_012_240_270
compacting	1_675_009_912	59_396_752	1_292_955_487	1_532_273_628	1_558_502_973
generational	2_517_025_054	59_397_016	977_578_942	1_052_786	967_410
incremental	32_320_741	4_624	290_257_785	292_951_006	292_977_552

Actor class

	binary size	put new bucket	put existing bucket	get
Map	254_076	638_613	4_449	4_909

Publisher & Subscriber

Measure the cost of inter-canister calls from the Publisher & Subscriber example.

	pub_binary_size	sub_binary_size	subscribe_caller	subscribe_callee	publish_caller	publish_callee
Motoko	139_886	126_827	14_641	8_451	10_530	3_662
Rust	472_135	519_916	51_591	34_661	74_169	41_615

* scale collection to 1M * 0.8M * fix and back to 1M * fix * disable heap * fix * add back heap * don't use heapify * uninstall canister to avoid CI OOM

github-actions · 2023-07-24T19:00:49Z

Note
Diffing the performance result against the published result from main branch.
Unchanged benchmarks are omitted.

Warning
Skip _out/collections/README.md, due to the number of tables mismatches from main branch.

Basic DAO

	binary_size	init	transfer_token	submit_proposal	vote_proposal
Motoko	225_805	37_517 ($\textcolor{red}{0.06\%}$)	16_312 ($\textcolor{red}{0.26\%}$)	12_696 ($\textcolor{red}{0.32\%}$)	14_153 ($\textcolor{red}{0.34\%}$)
Rust	704_886	471_865	86_525 ($\textcolor{red}{0.06\%}$)	104_617	115_765

DIP721 NFT

Note
Same as main branch, skipping.

Statistics

binary_size: no change
max_mem: no change
cycles: 0.21% [0.08%, 0.34%]

Heartbeat

	binary_size	heartbeat
Motoko	118_909	7_392
Rust	23_699	474 ($\textcolor{green}{-40.08\%}$)

Timer

Note
Same as main branch, skipping.

Statistics

binary_size: no change
max_mem: no change
cycles: no change

Warning
Skip table 0 ## Garbage Collection from _out/motoko/README.md, due to table shape mismatches from main branch.

Actor class

Note
Same as main branch, skipping.

Statistics

binary_size: no change
max_mem: no change
cycles: no change

Publisher & Subscriber

	pub_binary_size	sub_binary_size	subscribe_caller	subscribe_callee	publish_caller	publish_callee
Motoko	139_886	126_827	14_641 ($\textcolor{red}{0.06\%}$)	8_451	10_530	3_662
Rust	472_135	519_916	51_591	34_661	74_169	41_615

Statistics

binary_size: no change
max_mem: no change
cycles: 0.06%

Overall Statistics

binary_size: no change
max_mem: no change
cycles: 0.18% [0.07%, 0.30%]

* use Nat32 for PRNG * fix * fix * avoid debug_show overhead * use Nat64 * reuse rand for remove * scale collection to 1M (#70) * scale collection to 1M * 0.8M * fix and back to 1M * fix * disable heap * fix * add back heap * don't use heapify * uninstall canister to avoid CI OOM

chenyan-dfinity added 3 commits July 21, 2023 18:12

use Nat32 for PRNG

3c80da4

fix

18e9e1e

fix

87a12b9

chenyan-dfinity added 3 commits July 21, 2023 21:49

avoid debug_show overhead

e122cef

use Nat64

89cf0d5

reuse rand for remove

4fe8a4e

chenyan-dfinity changed the title ~~use Nat32 for PRNG~~ use Nat64 for PRNG Jul 23, 2023

kentosugama approved these changes Jul 24, 2023

View reviewed changes

scale collection to 1M (#70)

c6278af

* scale collection to 1M * 0.8M * fix and back to 1M * fix * disable heap * fix * add back heap * don't use heapify * uninstall canister to avoid CI OOM

Merge branch 'main' into fix-random

9388b89

chenyan-dfinity merged commit c37c1df into main Jul 24, 2023
1 check passed

chenyan-dfinity deleted the fix-random branch July 24, 2023 21:01

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

use Nat64 for PRNG #69

use Nat64 for PRNG #69

chenyan-dfinity commented Jul 22, 2023 •

edited

Loading

github-actions bot commented Jul 22, 2023 •

edited by chenyan-dfinity

Loading

github-actions bot commented Jul 22, 2023 •

edited

Loading

github-actions bot commented Jul 24, 2023 •

edited

Loading

use Nat64 for PRNG #69

use Nat64 for PRNG #69

Conversation

chenyan-dfinity commented Jul 22, 2023 • edited Loading

github-actions bot commented Jul 22, 2023 • edited by chenyan-dfinity Loading

Map

Priority queue

MoVM

Statistics

Basic DAO

DIP721 NFT

Statistics

Heartbeat

Timer

Statistics

Garbage Collection

Actor class

Statistics

Overall Statistics

github-actions bot commented Jul 22, 2023 • edited Loading

Collection libraries

💎 Takeaways

Map

Priority queue

Sample Dapps

Basic DAO

DIP721 NFT

Heartbeat / Timer

Heartbeat

Timer

Motoko Specific Benchmarks

Garbage Collection

Actor class

Publisher & Subscriber

github-actions bot commented Jul 24, 2023 • edited Loading

Basic DAO

DIP721 NFT

Statistics

Heartbeat

Timer

Statistics

Actor class

Statistics

Publisher & Subscriber

Statistics

Overall Statistics

chenyan-dfinity commented Jul 22, 2023 •

edited

Loading

github-actions bot commented Jul 22, 2023 •

edited by chenyan-dfinity

Loading

github-actions bot commented Jul 22, 2023 •

edited

Loading

github-actions bot commented Jul 24, 2023 •

edited

Loading