On Sun, Feb 20, 2011 at 9:23 AM, Arnd Bergmann [off-list ref] wrote:

The description of the test case is probably suboptimal. What this does
is 32 KB accesses, with 32 KB alignment in the pre and post case, but 16 KB
alignment in the "on" case. The idea here is that it should never do
any access with less than "--blocksize" aligment.

Now I feel slightly confused :(.

-b 16384 implies blocksize = 16384, maxalign is 8mb due to count 32,

               ret = time_rw_interval(dev, count, pre, blocksize,
                                       align - blocksize, maxalign,
                                       do_write);   //
<----------------- read 16k at align - 16k with 8mb intervals?
                returnif(ret);

                ret = time_rw_interval(dev, count, on, blocksize,
                                       align - blocksize / 2, maxalign,
                                       do_write);  //
<----------------- read 16k at align - 8k with 8mb intervals?
	        returnif(ret);

                ret = time_rw_interval(dev, count, post, blocksize,
 	                               align, maxalign, do_write); //
<-------- read 16k@align with 8mb intervals?
		returnif(ret);

I hope I'm not missing something obvious...

This is what I think happens:
Since the partition is over 64 MB size and it can have 7 4 MB allocation units open,
writing to 8 locations on the drive separated 8 MB causes it to do garbage collection
all the time for 32KB accesses and larger. However, the "on" measurement is only
16 KB aligned, so it goes into T's buffer A for small writes, and does not hit
the garbage collection all the time, so it ends up being a lot faster.

Can't go to A. A is 8KB big. Strange...

A