> On Fri, Oct 4, 2013 at 10:01 PM, Ryan Press <[email protected]> wrote:
>
> I'm running 3.11 kernel, and I've been playing around with bcache. I'd like
> to set something up where I can spin down the hard disks to save power,
> reading and writing to the SSD as much as possible. It is working quite well
> performance wise, with normal sequential speed as well as greatly improved
> random access speeds.
>
> To avoid spinning up the disk for writes, I have enabled writeback mode. I
> would like to increase writeback_delay, but whatever I set it to has no
> effect. If I spin down the hard disk with 'hdparm -Y /dev/sda', touch a file
> on the mounted share, exactly 30 seconds after I touch the file (the default
> writeback_delay) the hard drive spins up, regardless of what I echo to
> writeback_delay.
>
> Any ideas? Do I need to set writeback_delay in some sequence for it to take
> effect?
Alright, after some more testing it seems that writeback_delay is
indeed working as intended. There is something else touching the
disk, and I'm pretty sure it's the btree stuff. It has 30 seconds
hardcoded in two different places as so:
if (btree_node_dirty(b))
queue_delayed_work(btree_io_wq, &b->work,
msecs_to_jiffies(30000));
...and...
if (!btree_node_dirty(b))
queue_delayed_work(btree_io_wq, &b->work, 30 * HZ);
I guess this commits the btree to both the SSD and the hard disk? Is
it possible to increase this delay without compromising integrity (say
for a power failure) or is this a major design change?
Thanks,
Ryan
--
To unsubscribe from this list: send the line "unsubscribe linux-bcache" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
