search

crypto-org-chain / cronos

Cronos is the first Ethereum-compatible blockchain network built on Cosmos SDK technology. Cronos aims to massively scale the DeFi, GameFi, and overall Web3 user community by providing builders with the ability to instantly port apps and crypto assets from other chains while benefiting from low transaction fees, high throughput, and fast finality.
Other
296 stars 238 forks source link

Problem: load generator might be the bottleneck in benchmark #1573

Closed yihuang closed 2 months ago

yihuang commented 2 months ago

the idea is to generate all the test accounts and all the cosmos transactions in the gen phase.

mmsqe commented 2 months ago

I think might due to limited CPU resources competing by docker containers, killing fullnodes when all txs sent, validators can utilize all CPUs and consume much faster.

block 6 7761 2024-09-16T12:02:31.343621261Z
block 7 7761 2024-09-16T12:02:34.409613679Z
block 8 7761 2024-09-16T12:02:37.342882791Z
block 9 7761 2024-09-16T12:02:40.521641585Z
block 10 7761 2024-09-16T12:02:43.757741378Z
yihuang commented 2 months ago

I think might due to limited CPU resources competing by docker containers

that should not happen in cluster testing where each pod runs in its own node, right?

yihuang commented 2 months ago

killing fullnodes when all txs sent, validators can utilize all CPUs and consume much faster.

in that case, enable validator-generate-load and not run full nodes at all, should produce maximum outcome?

mmsqe commented 2 months ago

killing fullnodes when all txs sent, validators can utilize all CPUs and consume much faster.

in that case, enable validator-generate-load and not run full nodes at all, should produce maximum outcome?

I'm not familiar with infra, but there're 16 cpu per node, with a few cpu resource allocated for system usage, so there're only ~15.9 cpu for workload

yihuang commented 2 months ago

killing fullnodes when all txs sent, validators can utilize all CPUs and consume much faster.

do you mean in local run or cluster run? or do you suggest in cluster, the nodes also share CPU resources?

mmsqe commented 2 months ago

killing fullnodes when all txs sent, validators can utilize all CPUs and consume much faster.

do you mean in local run or cluster run? or do you suggest in cluster, the nodes also share CPU resources?

Cluster, my local only have 12 cores, and there're enable and disable validator-generate-load result, both have pros and cons , but cpu usage for validator is small.

yihuang commented 2 months ago

killing fullnodes when all txs sent, validators can utilize all CPUs and consume much faster.

do you mean in local run or cluster run? or do you suggest in cluster, the nodes also share CPU resources?

Cluster, my local only have 12 cores, and there're enable and disable validator-generate-load result, both have pros and cons , but cpu usage for validator is small.

I see, need to check with devops then, I thought they are independent.