[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [microblaze-uclinux] Performance microblaze vs. VirtexIIPro
Ensuring realtime synchrony is a top priority for embedded devices.
I/O, including TCP/IP, as well as the realtime clock itself, could be
"factored out" of the MB/uCL sequential kernel, and replaced with TCP/IP
& RTC device gate arrays operating in parallel with the rest of the
MB/uCL system. The kernel would call software hooks into those devices
instead of sequentially executing the original kernel instructions. That
would be a major win for the MB/uCL architecture over ASIC/uC
architectures, and point the way for other factored parallelism.
On Wed, 2004-05-12 at 02:19, John Williams wrote:
> Hi Josef, emanuel,
> jozsef imrek wrote:
> > On Tue, 11 May 2004, emanuel stiebler wrote:
> >>Anaybody out here made some tests what the performance in
> >>dhrystones is of this "chips" ?
> > there is a config option in the uClinux-dist that you can use to have a
> > dhrystones test. i switchd it on, run the test (the header says: Dhrystone
> > Benchmark, Version 2.1 (Language: C) ).
> > the fpga is a xc2v1000fg456-4 runing at 66Mhz.
> > at boot time the kernel measures 3.19 BogoMIPS.
> You can multiply this bogoMIPs by a factor of 10 if you enable the
> instruction cache. dcache has a very small effect, maybe <1 bogoMIPs.
> > the result is 5035.0 dhrystones/sec (see attached output from the test).
> Similarly the dhrystone measures are greatly improved if you enable the
> icache - without it, and running from external DDR, poor little
> microblaze lives in a perpetual pipeline stall.
> A note on the dhrystones, and an item for the wishlist - there is a
> problem with time accounting in the kernel - if you run dhrystone under
> zero load, the system clock time reported for execution of the benchmark
> is pretty accurate (compared to wall clock time).
> If you place the kernel under a ping flood attack the reported system
> time is unchanged, but the actual wall clock time doubles!
> I've never looked further into it, but it certainly should be
> straightened out, especially if you are relying on the system timer to
> even vaguely represent true time. This may be a common problem, which
> could explain why people put RTC chips in their systems if it really
> matters. Comments?
> microblaze-uclinux mailing list
> Project Home Page : http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux
> Mailing List Archive : http://www.itee.uq.edu.au/~listarch/microblaze-uclinux/
(C) Matthew Rubenstein
microblaze-uclinux mailing list
Project Home Page : http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux
Mailing List Archive : http://www.itee.uq.edu.au/~listarch/microblaze-uclinux/