benchmark: use much smaller values for n in some http tests - #14002
benchmark: use much smaller values for n in some http tests#14002psmarshall wants to merge 1 commit into
Conversation
Sorry, something went wrong.
|
No comment on the changes themselves, but the commit subsystem should be benchmark, not test. |
Sorry, something went wrong.
|
@cjihrig Thanks, I've updated the subsystem. |
Sorry, something went wrong.
BridgeAR
left a comment
There was a problem hiding this comment.
I just checked the benchmarks and with these low numbers the benchmark is not fully limited by the CPU and the numbers change to what they would be with higher n.
It will depend on the CPU but I would not recommend to go much below 5e7. I normally always change n when I run benchmarks locally depending on if I want to have a high accuracy or a glimpse at the likely outcome.
Sorry, something went wrong.
|
Ping @psmarshall would you mind updating the numbers once more? |
Sorry, something went wrong.
|
I decided to land this as is even with my comment. We can always increase the numbers when running the benchmarks locally if needed. Landed in 8a968e4 |
Sorry, something went wrong.
PR-URL: #14002 Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de> Reviewed-By: James M Snell <jasnell@gmail.com>
PR-URL: #14002 Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de> Reviewed-By: James M Snell <jasnell@gmail.com>
PR-URL: nodejs/node#14002 Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de> Reviewed-By: James M Snell <jasnell@gmail.com>
PR-URL: nodejs/node#14002 Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de> Reviewed-By: James M Snell <jasnell@gmail.com>
PR-URL: #14002 Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de> Reviewed-By: James M Snell <jasnell@gmail.com>
PR-URL: #14002 Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de> Reviewed-By: James M Snell <jasnell@gmail.com>
Checklist
Affected core subsystem(s)
benchmark
The n values here are really, really high right now, and we can still get high confidence (3 stars) with this much lower n value. This makes the benchmarks run a lot faster without sacrificing accuracy.