Bill Holler wrote:
> We currently do not allow the user to select a power/performance
> profile. This seems like something we eventually require for
> tunablity. 
> 
> The initial OpenSolaris could not introduce any performance
> regressions over a wide range of benchmarks.  There is a huge
> opportunity for 
> tuning for power if we are not constrained by 100% performance.

Should we loosen constrainst (of the 100% performance) for micro benchmarks 
like libmicro? For examples: we see 2~3x speedup of some microbenchmarks in 
libmicro with optimized string functions, and they actually bring 1.02x~1.03x 
in synthesis benchmarks. So 10% or 20% regression or improvement in 
microbenchmark can even bring non-visible impact to synthesis benchmarks or 
user workloads. I suggest we use user visible benchmark to make tradeoffs 
between power and performance.

- Frank


Reply via email to