bench get_retransmit_peers #23292
bench get_retransmit_peers #23292
Conversation
Codecov Report
@@ Coverage Diff @@
## master #23292 +/- ##
========================================
Coverage 81.3% 81.3%
========================================
Files 572 572
Lines 155876 156085 +209
========================================
+ Hits 126815 127025 +210
+ Misses 29061 29060 -1 |
core/benches/cluster_nodes.rs
Outdated
| let mut rng = rand::thread_rng(); | ||
| let (nodes, cluster_nodes) = make_cluster_nodes(&mut rng); | ||
|
|
||
| let mut shred_seed = [0u8; 32]; |
There was a problem hiding this comment.
Does it make sense to generate a different seed on every iteration to test different behaviors?
There was a problem hiding this comment.
I've changed this slightly to more accurately represent the seed generation for each case and to include multiple shreds in the calculation. I didn't want to include random generation of the inputs to get_retransmit_peers_* measurements.
core/src/cluster_nodes.rs
Outdated
| if rng.gen_ratio(1, 7) { | ||
| None // No stake for some of the nodes. | ||
| } else { | ||
| Some((node.id, rng.gen_range(0, 20))) | ||
| } |
There was a problem hiding this comment.
Out of curiosity, how does this ratio of unstaked:staked nodes of 1:6 affect the performance of the shuffle? Does more or less unstaked/staked nodes make perf faster?
There was a problem hiding this comment.
Added some tests for comparison:
running 4 tests
test bench_get_retransmit_peers_compat_unstaked_ratio_1_2 ... bench: 1,679,677 ns/iter (+/- 2,628)
test bench_get_retransmit_peers_compat_unstaked_ratio_1_32 ... bench: 1,690,616 ns/iter (+/- 3,048)
test bench_get_retransmit_peers_deterministic_unstaked_ratio_1_2 ... bench: 1,478,714 ns/iter (+/- 13,556)
test bench_get_retransmit_peers_deterministic_unstaked_ratio_1_32 ... bench: 2,476,904 ns/iter (+/- 39,032)
test result: ok. 0 passed; 0 failed; 0 ignored; 4 measured; 0 filtered out; finished in 5.22s
core/src/cluster_nodes.rs
Outdated
| @@ -39,7 +43,8 @@ enum NodeId { | |||
| Pubkey(Pubkey), | |||
| } | |||
|
|
|||
| struct Node { | |||
| // pub for bench | |||
There was a problem hiding this comment.
If someone removes pub, the code will not compile; so the comment will be redundant.
Similarly other places below.
core/src/cluster_nodes.rs
Outdated
| // pub for bench | ||
| pub fn make_cluster<R: Rng>( |
There was a problem hiding this comment.
maybe rename this to make_rand_cluster to indicate this is a test/simulation only function.
There was a problem hiding this comment.
changed to make_test_cluster
mergify
bot
dismissed stale reviews from behzadnouri and carllin
Pull request has been modified.
Problem
No bench for baseline get_retransmit_peers
Summary of Changes
Refactor to add benches for get_retransmit_peers_compat and get_retransmit_peers_deterministic
Fixes #