Steve (May 14 2025 at 15:25): Steve (May 14 2025 at 15:25):I'm currently trying to use WAMR in a heavily parallelized application. I have benchmarks running for this application that shows that WAMR is hitting 86% of my native speed in a single thread but that performance drops off significantly with multiple threads.
I'm looking for advice from WAMR experts on how to build and use WAMR most effectively given my environment.
To clarify, I am running as many small WebAssembly modules as possible looking to optimize for time rather than space.
I'm currently building with fast_jit and without hardware bounds checks.
Steve (May 14 2025 at 15:25):------------------------------------------------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
------------------------------------------------------------------------------------------------------------------------------------------
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:1 47160 ns 46890 ns 148459 transaction_time=47.1604us transactions=21.2042k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:2 52020 ns 49768 ns 134234 transaction_time=26.0102us transactions=38.4465k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:4 64209 ns 57525 ns 111384 transaction_time=16.0522us transactions=62.2969k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:8 102226 ns 83479 ns 74656 transaction_time=12.7782us transactions=78.2583k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:16 225669 ns 156788 ns 32352 transaction_time=14.1043us transactions=70.9003k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:32 507825 ns 141228 ns 12896 transaction_time=15.8695us transactions=63.0139k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:64 896231 ns 154784 ns 8448 transaction_time=14.0036us transactions=71.4102k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:128 1613972 ns 158502 ns 4224 transaction_time=12.6092us transactions=79.3074k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:256 2915854 ns 157127 ns 3328 transaction_time=11.3901us transactions=87.7959k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:512 6507146 ns 183015 ns 1536 transaction_time=12.7093us transactions=78.6827k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:1024 8599593 ns 202500 ns 1024 transaction_time=8.39804us transactions=119.075k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:2048 24181246 ns 233584 ns 2048 transaction_time=11.8072us transactions=84.6937k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:4096 37847957 ns 211060 ns 4096 transaction_time=9.24022us transactions=108.222k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:8192 72479376 ns 206789 ns 8192 transaction_time=8.84758us transactions=113.025k/s
token_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:16384 142267166 ns 208227 ns 16384 transaction_time=8.6833us transactions=115.164k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:1 42141 ns 42085 ns 164714 transaction_time=42.141us transactions=23.7299k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:2 42439 ns 42378 ns 165524 transaction_time=21.2197us transactions=47.1261k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:4 45792 ns 45711 ns 150484 transaction_time=11.448us transactions=87.3519k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:8 56366 ns 56271 ns 124280 transaction_time=7.04573us transactions=141.93k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:16 83458 ns 83232 ns 85280 transaction_time=5.2161us transactions=191.714k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:32 151053 ns 106136 ns 41632 transaction_time=4.72042us transactions=211.846k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:64 249221 ns 114625 ns 41408 transaction_time=3.89408us transactions=256.8k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:128 457146 ns 113368 ns 15232 transaction_time=3.57146us transactions=279.998k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:256 799427 ns 108666 ns 9216 transaction_time=3.12276us transactions=320.229k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:512 1670683 ns 110192 ns 3072 transaction_time=3.26305us transactions=306.461k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:1024 3584987 ns 113055 ns 2048 transaction_time=3.50096us transactions=285.636k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:2048 9674782 ns 127742 ns 2048 transaction_time=4.72401us transactions=211.684k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:4096 19895501 ns 141395 ns 4096 transaction_time=4.8573us transactions=205.876k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:8192 38847796 ns 136568 ns 8192 transaction_time=4.74216us transactions=210.874k/s
coin_transactions/min_time:5.000/min_warmup_time:1.000/real_time/threads:16384 79561378 ns 134940 ns 16384 transaction_time=4.85604us transactions=205.929k/s
As you can see here, 21.2042k/s on WASM and 23.7299k/s natively. These numbers are really respectable. However as the threads increase native tops out at 320.223k/s while WAMR tops out at 119.075k/s.
Token = WASM
Coin = Native
lum1n0us (May 26 2025 at 01:14):I am not sure if my advice will fit your cases until I learn more details. Here are some common suggestions:
lum1n0us (Jun 23 2025 at 00:10):It seems like there hasn't been any activity on this issue for a while. If you have any further questions or updates, please feel free to share them with us.
To keep our project organized, we plan to close this issue soon if there are no new developments. Of course, you can always reopen it or start a new thread if needed.
Thank you for your understanding and cooperation!
Last updated: Dec 06 2025 at 07:03 UTC