[Libdlo] [RFC 2.6.30 1/1] defio fbdev displaylink implementation

[Phil Endecott]

Bernie Thompson wrote:
> One interesting area of performance concern for Jaya's driver (once
> compression is added) is the 4KB granularity of defio - sending any
> unnecessary pixels over USB will hasten our most important bottleneck.

Hi Bernie,

In my experience of similar problems in other domains, you have two cases:

- Small changes, where the 4k block size is inefficient; in this case 
it doesn't really matter that it's inefficient, because the changes are 
small anyway.

- Large changes, where the 4k block size is fine.

> 2) Enhance defio to provide "shadow page" support, such that when 
> a PTE is touched and backed, we also alloc a second page and take 
> snapshot of the soon-to-be-dirty page before the client has changed 
> it. Then we pass both before/after pages to the defio-supporting 
> driver to do a more precise detection of differences.

The question is, in the case where the whole page is going to be 
re-written (e.g. playing video), how much effort have you wasted doing 
the copy and comparison?  Changes that optimise for the case that was 
fast enough already, and pessimise the case that was already bad, are 
not a good choice.


Regards,  Phil.