Bug#702256: EnhanceIO redundant with bcache?
On Sat, May 04, 2013 at 11:17:36AM +1000, Dmitry Smirnov wrote:
> On Sat, 4 May 2013 04:45:00 Ben Hutchings wrote:
> > According to <https://lwn.net/Articles/548348/>, bcache is going into
> > Linux 3.10 (though it now looks like it has missed that). With
> > dm-cache and bcache in the kernel, would there still be a need for
> > this third option?
>
> Good question. :)
>
> I didn't have a chance to examine dm-cache yet but I tried others.
>
> To me bcache is the most worthless of all because it requires
> specially prepared (formatted) data partition. This makes it difficult
> (if possible) to attach cache to existing partition with data as one
> would need 200% capacity and to perform long data-moving in order to
> activate/deactivate caching. I don't know if this fundamental
> show-stopper changed since I tried bcache. Besides bcache also
> requires modified kernel which is limiting its "field of endeavour"
> even further.
But it won't require a modified kernel in future. That's my point.
> We packaged flashcache when there were no other feasible alternatives.
> To use flashcache it is necessary to create a virtual device-mapper
> device that combines data and cache devices. Assuming that data
> partition already have data it is only needed to mount it through
> associated flashcache device.
>
> The brilliance of EnhanceIO is that it doesn't need intermediate
> device at all and can be attached to any block device on-the-fly even
> when device is already mounted.
[...]
Bcache can apparently do that as well, but as you noted the backing
device must have been prepared with a bcache superblock. Apparently
the reasoning for this is to prevent accidental access that bypasses
the cache. It seems like this could be relaxed if the cache is
configured for write-through behaviour, though.
Ben.
--
Ben Hutchings
We get into the habit of living before acquiring the habit of thinking.
- Albert Camus
Reply to: