benchmark: add BenchmarkBlockEncoding #4590
Conversation
Codecov Report
@@ Coverage Diff @@
## master #4590 +/- ##
==========================================
- Coverage 54.47% 54.44% -0.04%
==========================================
Files 407 407
Lines 52389 52389
==========================================
- Hits 28541 28522 -19
- Misses 21460 21474 +14
- Partials 2388 2393 +5
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
| @@ -636,3 +639,52 @@ func callAppTransaction( | |||
| appTx.Type = protocol.ApplicationCallTx | |||
| return | |||
| } | |||
|
|
|||
| // BenchmarkBlockEncoding builds a full block of pay transactions and benchmarks the time it takes to encode b.N of them. | |||
| func BenchmarkBlockEncoding(b *testing.B) { | |||
There was a problem hiding this comment.
misnamed? This is really a benchmark of zstd compression at some level? I was expecting a benchmark around the msgp operation as 'encoding'
There was a problem hiding this comment.
Oh right, it was benchmarking the encoding before, and then I moved the place where ResetTimer and StopTimer was surrounding when Pavel started implementing block compression
| currentRound := bc.round | ||
| for bc.round == currentRound { | ||
| // add pay transaction | ||
| //payEvent(bc, mrand.Float64() < newAcctProb) |
| //newAcctProb := 0.0 | ||
| //newAcctProb := 1.0 | ||
| bc := setupEnv(b, numAccts) | ||
| zstdLevel, err := strconv.Atoi(os.Getenv("ZSTD_LEVEL")) |
There was a problem hiding this comment.
who sets this env variable?
| payEvent(bc, false) | ||
| } | ||
| if (currentRound+1)*10%(2*numBlocks) == 0 { | ||
| fmt.Printf("%d%% %.1fs ", (currentRound+1)*100/numBlocks, time.Since(s3).Seconds()) |
There was a problem hiding this comment.
nit: use b.Logf or fmt.Printf. Mixing them is confusing.
| b.ResetTimer() | ||
| for i := range encs { | ||
| zstdbufs[i], err = zstd.CompressLevel(nil, encs[i], zstdLevel) | ||
| b.Logf("zstd compress size: %d orig: %d", len(zstdbufs[i]), len(encs[i])) |
There was a problem hiding this comment.
This is not very useful (i.e. printing a line for each). Will be nice to collect the stats are print at the end as a summary of mean/median of the compression ratio.
Also, this may also impact the accuracy of the time measurement.
|
This was helpful but no longer needed |
Summary
This is a simple benchmark, built on Shant's block validation benchmark support code in #4522, to benchmark the encoding time of a full block.
Test Plan
No changes except this additional benchmark.