Experiment different javascript runtimes

[twoeths]

Problem description

So far we only run lodestar with NodejS. The goal is to experiment other javascript runtimes like Bun and Deno to see if lodestar performs better there. I had different benchmark for them:

• Memory: as tested [here](url), Bun is very efficient to store a short Uint8Array of 32 bytes which is critical for our merkle tree. This could help unblock Reducing CachedBeaconState size #5857

Number of bytes used to store an Uint8Array of 32 bytes

-	NodeJS 22.4.1	Bun 1.1.34	Deno 2.0.5
Uint8Array(32)	241.74	47.35	200.65

• Performance - deserialize a BeaconBlock:
  • deserialize a BeaconBlock of 200 transactions, 500 bytes each: https://github.com/ChainSafe/ssz/blob/d7417022d15755a70b481f6f815ae08a4e685cec/packages/ssz/deserialize_beacon_block_perf_test.ts#L6
  • modify 100k validators of a 2M-validator merkle tree: https://github.com/ChainSafe/ssz/blob/d7417022d15755a70b481f6f815ae08a4e685cec/packages/persistent-merkle-tree/nodejs_bun_deno.ts#L5

Here's my result on a Mac M1:

-	NodeJS 22.4.1	Bun 1.1.34	Deno 2.0.5
Deserialize a BeaconBlock	0.25ms	0.11ms	0.719ms
Modify 100k nodes of 2M merkle tree	23.86ms	4.727ms	15.889ms

to run the NodeJS, I used the tsx and it's worth to note by @nazarhussain that Typescript overhead is almost negligible in case of Bun, ~10% in case of Deno and 100% in case of TSX/TS-Node

Solution description

Based on the test results I'd like to discuss how to go from there

• Bun is potentially to be a candidate javascript runtime that help boost lodestar's performance, do we need other tests for it?
• One blocker for Bun is the lack of napi support, how to deal with that? Implement Node-API in bun.js (napi) oven-sh/bun#158

Additional context

No response

[twoeths]

also note the result run by @nazarhussain here:

For such comparisons we should avoid typescript complexity. I run it over the build JS files and got different results. But yet Bun is the winner. 

Node: 9.1ms 
Deno: 8.2
Bun: 3.71

One other observation is that Typescript overhead is almost negligible in case of Bun, ~10% in case of Deno and 100% in case of TSX/TS-Node

[ensi321]

Looks like Bun's performance trumps Deno's by a substantial amount. Is there any advantage of Deno over Bun that we should consider?

[twoeths]

Looks like Bun's performance trumps Deno's by a substantial amount. Is there any advantage of Deno over Bun that we should consider?

@ensi321 I don't see any, as discussed in the previous meeting we want to explore Bun because its JavaScriptCore engine is way different from NodeJS + Deno. One obstacle for Bun is it does not support Node API, we should try Bun's FFI instead and compare the performance with Node API

[wemeetagain]

According to https://bun.sh/docs/api/node-api (and verified doing some basic tests), bun supports most of NAPI