[Barrelfish] [Barrelfish-users] System knowledge
Michael N. Gagnon
mgagnon at seas.harvard.edu
Wed Oct 7 00:52:21 MEST 2009
Hi again, and thank you for the quick reply. I have another question.
I took a quick look at the source code and it looks like there are
user-mode and kernel-mode micro benchmarks.
kernel/include/kernel/microbenchmarks.h defines MICROBENCH_ITERATIONS to
usr/bench/bench.c defines MICROBENCH_ITERATIONS to be 1000000
I am curious, what motivated these particular values?
Timothy Roscoe wrote:
> Dear Mike,
> We do indeed perform microbenchmarks on startup - for example, the
> multicast tree mentioned in the SOSP paper is constructed using an SKB
> query over core-core messaging latency values measured when Barrelfish
> first boots. Most of the current online microbenchmarks are driven by
> the immediate needs of the OS, but we certainly plan to extend these in
> the future.
> Best regards,
> -- Mothy
> Michael N. Gagnon wrote:
>> Hi Barrelfishers,
>> "Embracing diversity in the Barrelﬁsh manycore operating system"
>> says that Barrelfish will use (or uses?) a "system knowledge base" to
>> educate system-resource policies and derives the SKB using "online
>> measurement and profiling." The paper specifically mentions application
>> benchmarking but does not explicitly mention the use of
>> micro-benchmarking. It seems micro-benchmarking would be useful to
>> measure properties of the hardware system (like X-Ray ). Does
>> Barrelfish use micro-benchmarks this way?
>>  Kamen Yotov, Keshav Pingali, and Paul Stodghill.
>> X-ray: A tool for automatic measurement of hardware
>> parameters. In International Conference on the
>> Quantitative Evaluation of Systems, 2005.
>> Thank you,
>> Michael N. Gagnon
>> Computer Science Ph.D. Student, Harvard University
>> Barrelfish-users mailing list
>> Barrelfish-users at lists.inf.ethz.ch
>> Barrelfish mailing list
>> Barrelfish at lists.inf.ethz.ch
More information about the Barrelfish-users