[dm-devel] New dm-bufio with shrinker API

[Joe Thornber]

On Tue, Sep 06, 2011 at 12:08:51PM -0400, Mikulas Patocka wrote:
> So, I can implement a call "void dm_bufio_discard_buffer(struct 
> dm_bufio_client *c, sector_t block)" and this call will discard a specific 
> buffer at a specific location (the call is not guaranteed to succeed, it 
> would not discard if someone is holding the buffer or so). It would be 
> like a "bforget" function for filesystem buffers.
> 
> You will call this function on metadata sectors that you are freeing.
> 
> Do you agree with this interface?

That's not really what I was after, but it's a good idea.  Don't do
anything about it now, and I'll instrument to see if I can drive it
effectively.  I try and recycle freed blocks as quickly as possible to
avoid fragmenting free space, which may mean there is little benefit.

> Other than this, I don't know how to reduce cache size, I don't know about 
> any algorithm that would guess cache size automatically. In operating 
> systems, caches usually grow without limit regardless of whether someone 
> needs the cache or not.

ok, let's go with things as they are.  Thx for your hard work.

One other optimisation to think about: As you know, if a non-blocking
lookup of the thinp mapping fails, the bio gets handed across to a
worker thread to do the blocking lookup.  Is there any way to you
could make dm_bm_read_try_lock() pass a preload hint to bufio, since
we know that block is going to be required in the near future?

- Joe