Performance

Performance and memory

Test date: 9/1/2020

X86-64 platform:

CPU：Intel(R) Core(TM) i7-7700 CPU @ 3.60GHz
OS ：ubuntu 18.04 64-bit
Compiler：gcc version 7.5.0 (Ubuntu 7.5.0-3ubuntu1~18.04), flag: O3

performance comparison:

workload	wamr-aot / wasm3	wasm3 / wamr-interpreter	wamr-aot / GCC(O3)	wamr interpreters (fast/classic)
matrix	22.02	1.23	0.68	2.47
gimli	9.94	1.07	0.94	2.08
CoreMark	8.79	1.32	0.79	2.52

Note：

• The numbers in the column are the performance ratios of the two runtime engines provided. Workloads matrix and gimli are measured by the execution time, and CoreMark is measured by its reported score. A/B denotes the execution_time_of_B/execution_time_of_A, or the score_of_A/score_of_B. For example, matrix 22.02 means that execution_time_of_wasm3/execution_time_of_wamr-aot is 22.02, CoreMakr 8.79 means score_of_wamr-aot/score_of_wasm3 is 8.79.
• WAMR provides two interpreters - the fast version for advanced performance and and the classic version for smaller footprint respectively.

ARMV7 Cortex-M7:

Board: nucleo_f767zi (ARMV7 Cortex-M7)
OS : Zephyr
Compiler: arm-zephyr-eabi-gcc (crosstool-NG 1.24.0.37-3f461da-dirty) 9.2.0, Os

performance comparison:

workload	wamr-aot/wasm3	wasm3/wamr-interpreter	wamr interpreters (fast/classic)
matrix	30.17	1.06	1.65
gimli	19.13	1.11	2.03
seqhash	21.12	1.15	1.34

Memory consumption

workload: CoreMark
measurement tool: valgrind/massif

WASM interpreter	Peak memory usage
WAMR classic	365KB
WAMR fast	485KB
wasm3	514KB

WASM application binary size

Compilation option: -Os

Workload	GCC compiled native binary (Bytes)	Wasm bytecode (Bytes)	iwasm-aot module (Bytes)
coremark	23120	10457	17348
base64	8792	1026	1840
gimli	8640	498	1080

Misc.

The page is also a response to the data published by wasm3 at https://github.com/wasm3/wasm3/blob/master/docs/Performance.md.

Some background of the performance comparison between WAMR and wasm3:

https://github.com/bytecodealliance/wasm-micro-runtime/issues/134