Re: [PATCH 2/2] Documentation: add description of dirty_background_time

[Fengguang Wu <fengguang.wu@intel.com>]

On Mon, Aug 27, 2012 at 12:01:41PM -0400, Namjae Jeon wrote:
> From: Namjae Jeon <namjae.jeon@samsung.com>
> 
> This commit adds dirty_background_time description in bdi sysfs
> documentation.
> 
> Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com>
> Signed-off-by: Vivek Trivedi <t.vivek@samsung.com>
> ---
>  Documentation/ABI/testing/sysfs-class-bdi |   13 +++++++++++++
>  1 file changed, 13 insertions(+)
> 
> diff --git a/Documentation/ABI/testing/sysfs-class-bdi b/Documentation/ABI/testing/sysfs-class-bdi
> index 5f50097..018e26a 100644
> --- a/Documentation/ABI/testing/sysfs-class-bdi
> +++ b/Documentation/ABI/testing/sysfs-class-bdi
> @@ -48,3 +48,16 @@ max_ratio (read-write)
>  	most of the write-back cache.  For example in case of an NFS
>  	mount that is prone to get stuck, or a FUSE mount which cannot
>  	be trusted to play fair.
> +
> +dirty_background_time (read-write)
> +
> +	It is used to start early writeback of given bdi once bdi dirty
> +	data exceeds product of average write bandwidth and
> +	dirty_background_time.

It may be worth to note that it works _in parallel with_ the global
background writeback threshold rather than replacing it.

> It is mainly useful for tuning writeback
> +	speed at 'NFS Server' so that NFS client could see better write speed.

Hopefully more clear: 

A good use case is setting it to around 100 (1 second) in the NFS
server for improving NFS write performance. Note that it's not
recommended to set it to a too small value, which might lead to
small IO size. Setting it to 0 disables the feature.

> +	However, sometimes it may not match user expectations as it is based
> +	on bdi write bandwidth estimation.

The users should not expect this threshold to work accurately.

> Write bandwidth estimation is a
> +	best effort to estimate bdi write speed bandwidth. But it can be
> +	wildly wrong in certain situations.

such as sudden change of workload (including the workload startup stage),
or if there are no heavy writes since boot, in which case there is no
reasonable estimation yet.

> +	dirty_background_time is expressed in msec.

That can be eliminated if changing to dirty_background_centisecs.

Thanks,
Fengguang