Vitaly,
On Thursday 05 October 2006 14:29, you wrote:
> > The fact that your numbers show that Yaffs is faster in the
> > nandsim test, indicates to me that the underlying
> > performance of the real NAND device and hook-up is a large
> > factor in any measured performance. nandsim really doesn't
> > simulate NAND.
>
> Well, first, nandsim does simulate NAND :)
The default delays in nandsim are a lot shorter than the delays
for the chips and buses we see. The LH79524 ARM, for example,
has a 'output cycle' of 158nS, nandsim has 40. This makes a
big difference with 2K page, with approx 60% of time being
transfer and 40% chip program.
> Then, I'm totally aware that this is not a comprehensive
> comparison. Still I think that we can make some conclusions
> based on the results, and the conclusions are that yaffs2 is
> slower than expected on write operations for 2k page flashes
> and slower on mount than expected on 512b flashes, and that's
> something we'd better profile.
There is a lot of value knowing the underlying performance of
NAND i/o on a platform and I agree with you that there is value
in comparing real NAND with nandsim and seeing what happens to
the filesystem performance -- what I didn't like about the
original posting was the lack of baseline facts like NAND i/o
rates/timing, nandsim parameters etc., without these I can't
deduce very much from the figures.
-imcd
(text/plain)