Re: [PATCH] mm: moving dirty pages balancing to pdfludh entirely

[Nikita Danilov <nikita@clusterfs.com>]

Ananiev, Leonid I writes:
 > Nikita Danilov writes:
 > > Exactly to the contrary: as I explained to you, if you have more
 > devices
 > > than pdflush threads
 > I do not believe that Bret Towe has more devices than
 > MAX_PDFLUSH_THREADS=8.

Some people do, should they suffer? :-)

 > 
 > > See how wbc.nr_to_write is set up by balance_dirty_pages().
 > It is number TO write but I said about number after what user has to
 > write-out all dirty pages. 

Not _all_, only nr_to_write of them:

			if (pages_written >= write_chunk)
				break;		/* We've done our duty */

 > 
 > > imagine that MAX_PDFLUSH_THREADS equals 1
 > Imagine that CONFIG_NR_CPUS=1 for smp.
 > Kernel has a lot of "big enough" constants.

Then why introduce more of them?

In current design each thread is responsible for write-out. This means
that write-out concurrency level scales together with the number of
writers. You propose to limit write-out concurrency by
MAX_PDFLUSH_THREADS. Obviously this is an artificial limit that will be
sub-optimal sometimes.

 > 
 > Leonid

Nikita.