General Performance Review Topics [re: Xeon Running FBSD]

grarpamp grarpamp at gmail.com
Sat Oct 14 02:27:02 UTC 2017


> To get more consistent results, you may actually want to disable turboboost in the bios, and rerun the benchmarks on ALL of the operating systems.

> Wow, the difference between the FreeBSDs and Linux performance is amazing and for those looking at the first time on such benchmarks not knowing much about the turbo boost issue one would definitely choose the faster one :-(

> I'd appreciate results of a benchmark considering "x"...

> AMDs new Epyc platform comprised also from lots of cpu cores

> You might want to try enabling turbo boost.
> sysctl dev.cpu.0.freq=9999

Not on the list so I could not see if you folks were CC'ing
the reviewers (since the freebsd list mbox archives fail to
include the important CC lines.). They certainly should be
informed of review feedback.

So are reviewers going to spend time on searching, mailing, and
applying all manner of specific tuning and recompiling trying to figure
out (and/or fix) why performance of OS 1 sucks compared to OS 2?

Alternatively, who is doing hardcore reviews of unix OS?
What is the demand for that?

Or are they going to take the released default boxed version
of any given set of OS and put it through its paces such as
any relatively newcoming user might do?

And thus why should, or should not, FreeBSD just
ship with freq=9999, or any other generic, obvious,
faq'd, or necessary tuning by default?

Are we worried about melting down somone's silicon
by shiping with turbo boost enabled?
About some bugs?
About maintaining balance of results across workloads?

Is there concern about accurate protrayal of long running
performance where cpu heats up thus slowing performance
down to steady state, vs short performance test runs of perhaps
less than a minute that some reviewers might be using?

Is there bilateral participation with reviewers on upcoming
review plans, methodologies, feedback, improvements, etc?

A structure / group in place within FreeBSD tasked with
identifying, developing, integrating performance improvements?

Thoughts.


More information about the freebsd-performance mailing list