5

During the last Substrate Builders Program Office Hours, the Mangata team raised awareness around some potential confusion around the Substrate Benchmarking System and how it is being applied in the Polkadot Ecosystem.

Polkadot Wiki suggests some Standard Hardware, which are the same as in the Substrate Docs.

Meanwhile, OnFinality Polkadot node hardware has modest 4GB of memory, and go even lower for parachain nodes.

How do we reconcile those differences? Can OnFinality's nodes hit bottlenecks earlier than expected, due to their deviation from standard hardware?

2 Answers 2

4

The hardware requirements that you found are correct.

There are multiple things to consider when thinking about hardware requirements.

Type Of The Node

The hardware requirements are aimed at Validators which participate in the consensus.
They need to execute time-critical calculations and can be slashed for not meeting the deadlines.
In order to have some baseline to calculating the deadlines, we opted for a reference hardware.
Ultimately this is only guideline and you can make your own choice.
Parity just provides these requirements as a reference, not a hard requirement.

Archive nodes and pruning nodes can basically use any hardware they see fit since they are not consensus relevant.

Custom indexer have to measure their own requirements.
Since most blocks are currently not full, you can probably use very cheap hardware.

Chain Usage

Polkadot blocks are mostly not even close to their maximum possible weight.
Since the reference hardware is designed for 100% chain usage, this translates to much less resource usage than to what is possible,

Regarding your concrete question:

Can OnFinality's nodes hit bottlenecks earlier than expected, due to their deviation from standard hardware?

Yes, it is possible that they would not keep up if the chain usage surges and they are using subpar hardware.
In this case its on them to make sure that this does not happen.

2
  • when you say "the hardware requirements are aimed at Validators which participate in the consensus.", would you include Collators in that category as well? Apr 13, 2022 at 19:10
  • Yes also for collators. Apr 26, 2022 at 10:33
4

relay validators execute parachain blocks... if parachain collators are benchmarked on HW that's faster than standard, that means relay validators will spend more time than expected executing parachain blocks, which will lead to issues on consensus...

honestly, I think that going with standard HW is just easier because it avoids confusion and potential footguns!

but if really needed, it should be ok if parachain runtimes are benchmarked so that collators have specs that are lower than standard, but not higher...

that would probably be a waste of relay blockspace though 🤷... probably not a big issue if there's not much throughput on the parachain, but definitely wasteful when parachain blocks are full!

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.