[RFC/PATCH 2/3] NAND multiple plane feature

[Alexey Korolev]

Hi,

> On Tue, 3 June 2008 17:57:35 +0100, Alexey Korolev wrote:
> > 
> > We did some performance testing for this feature: we did not find a test
> > case where JFFS2 on NAND with dual-plane has lower performance than
> > JFFS2 on NAND without dual-plane features. (note: subpage read feature
> > was enabled for both cases)
> 
> Maybe I have a testcase for you.  The "evil workload".
> 
> 1. Write a file until you get -ENOSPC.
> 2. Replace 1000 page-sized page-aligned blocks at random offsets.
> 
> The time for 2 should roughly double with bigger eraseblocks.
> 
> [ And yes, I originally got this testcase from you. ;) ]
We just tried to check your scenario. It is not
possible to overwrite something if JFFS2 returned -ENOSPC until
something got deleted. 

In fact our test suite do not include this test case. We have another.
Write file1 at about 5% of whole space, write file2 for rest of space.
Delete file1. Do step 2. 
In this not so good case we got the following numbers:

single-plane:
73 sec 424541 usec

dual-plane:
69 sec 428637 usec

Thanks,
Alexey