twoeths

> Yes I agree this one is tricky. We wanted to try and modify the scoring system to not use this parameter. 👍 for this > The point of the...

the bottom up view shows that `gauge.inc()` takes 2.45% of cpu time which is too much for a comparison, `runHeartBeat` takes 3.6% (before the improvements)

> Check if performance can be improved for Gauge.inc(): harder @dapplion `gauge.inc()` is slower than `counter.inc()` because it hash `labels` object 2 times, see https://github.com/siimon/prom-client/pull/503

since the `hashObject()` in `prom-client` is not cheap, if we could cache metric value and flush every n times or per scraping metric ```typescript class LabelCachedGauge { private twoLabelsCache: Map...

thanks @achingbrain , protons 5.1.0 is really faster than older versions however it's not very stable, see the statistics here https://github.com/ChainSafe/js-libp2p-gossipsub/pull/327

> Also, how are you generating those stats, I can't see anything new in the benchmark folder? I just run the test in my local environment > That's interesting, [email protected]...

I have no issue running the benchmark ``` npm run benchmark > @chainsafe/[email protected] benchmark > node ./node_modules/.bin/benchmark 'dist/test/benchmark/*.test.js' --local Connected to historyProvider: LocalHistoryProvider, dirpath ./benchmark_data No previous bencharmk found for...

with new buckets `[0.01, 0.05, 0.1, 0.2, 0.4, 0.7, 1]`, the distribution in a contabo-S (with rate interval as 5m) looks like

was able to catch a 775ms heartbeat, it shows that most of the time of `heartbeat()` function is due to `emitGossip()` call [0522_contabo_20_subscribe_all_subnets.cpuprofile.zip](https://github.com/ChainSafe/js-libp2p-gossipsub/files/8820289/0522_contabo_20_subscribe_all_subnets.cpuprofile.zip) right now we call `emitGossip()` 2 times...

most of the cost is for looping through topic peers and for a lodestar node connecting to all attestation subnets, there are about 70 topics in total => we call...