Improve sTPS calculation (ignore non-call txs, ignore last block) #58
Conversation
|
Now the results look much better, LGTM |
src/stats.rs
Outdated
| println!("TPS - Transaction execution time per second, assuming a 0.5-second execution time per block"); | ||
| println!( | ||
| "TPS: {}", | ||
| tps_total_extrinsics as f64 / (tps_blocks as f64 * 0.5) |
There was a problem hiding this comment.
With async_backing we should have a 2s execution time now. Not sure if the target test bins have been updated to a version which has that enabled or not.
There was a problem hiding this comment.
given currently used zombienet configuration for launching polkadot relay chain, I can see in the resulting/output chain configuration that async is disabled (see example below)
cat /tmp/zombie-1d78b25178b5f5a041c1558f57a34514_-123208-MoKciO8z6eTZ/cfg/rococo-local-plain.json | grep -A5 async
"async_backing_params": {
"allowed_ancestry_len": 0,
"max_candidate_depth": 0
},
"code_retention_period": 1200,
"coretime_cores": 0,
cat /tmp/zombie-1d78b25178b5f5a041c1558f57a34514_-123208-MoKciO8z6eTZ/cfg/rococo-local-plain.json | grep scheduling
"scheduling_lookahead": 1,
There was a problem hiding this comment.
Moonbeam did not support async backing until last release, which was made last week.
I think that first we need to make the flow working and after this we can adjust graphs and configuration.
WDYT?
There was a problem hiding this comment.
Now we calculate sTPS using block build time. Async backing will be delivered with this PR: #60
smiasojed
changed the title
3 participants
The sTPS calculation is based on the following assumptions about smart-bench:
sTPS calculation is exclusively concerned with contract calls, disregarding any system or contract-creating transactions.
sTPS calculation excludes the last block of the benchmark, as its full utilization is not guaranteed. In other words, only blocks in the middle will consist entirely of contract calls.