bench: increase benchmark throughput data size #5587
Conversation
kaukabrizvi
changed the title
| pub fn bench_throughput_cipher_suites(c: &mut Criterion) { | ||
| // arbitrarily large to cut across TLS record boundaries | ||
| let mut shared_buf = [0u8; 100000]; | ||
| let mut shared_buf = vec![0u8; 10_000_000]; // 10MB (allocated on the heap to avoid stack overflow) |
There was a problem hiding this comment.
nit: let's move this to a top-level comment instead of having both top-level comment and inline comment describing the same line
| pub fn bench_throughput_cipher_suites(c: &mut Criterion) { | ||
| // arbitrarily large to cut across TLS record boundaries | ||
| let mut shared_buf = [0u8; 100000]; | ||
| // 10MB buffer - arbitrarily large to cut across TLS record boundaries. | ||
| let mut shared_buf = vec![0u8; 10_000_000]; |
There was a problem hiding this comment.
Is this the right 10MB? I'm not super familiar with the benchmark tests, and I also see
Did you just test that all benchmarks passed, or did you somehow confirm that all benchmark tests now send more data / take longer? Can you expand your "Testing" section with more details about how you know this change achieves what you're trying to achieve?
There was a problem hiding this comment.
This is the right 10 MB to configure buffer size for throughput benchmarks. To address the second point, I inspected the benchmark results more closely and found that s2n-tls performs significantly faster than expected at 10 MB. I'm currently exploring this and will follow up once I've gathered sufficient data and have more clarity on this behavior.
3 participants
Goal:
Increase throughput test size from 100 KB to 10 MB for more representative benchmarking.
Why:
To smooth out graphs in the Ops dashboard.
How:
Expanded shared buffer size and moved allocation to the heap to prevent stack overflow.
Testing:
Validated locally via cargo bench; all benchmarks passed.
Next Steps:
Update the Ops dashboard scale and documentation to reflect the new throughput size.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.