Xen-blkback bugs/features

Xen-blkback bugs/features

[Konrad Rzeszutek Wilk]

Hey Roger,

Since you are taking a look at blkback and continuing the work
that Oliver did with persistent grants (and btw, thanks for doing
that!) I was wondering if you were also looking at some of the
"todos" that were circulating around.

One of them was the 4KB issue. If one hooks up bcache as backend
into blkback, the I/Os don't seem to go anywhere. Both James Harper
and .. I think the author of bcache were talking about this but
it never got anywhere.

The other one is expanding the segment limit. We have hard-coded
it to be 11 but with existing use-cases (say SSDs) that is not
enough. Intel had posted an RFC patchset that created a new ring
for segments. SpectraLogic (Justin) did it by expanding the ring.

Then there is the prototype work that Martin Peterson (really really
prototype) to get DIF/DIX plumbed in xen-blkback/xen-blkfront. He
needed 8 bytes per bio so he ended up creating a new ring to track that.

I think those are the outstanding use-cases that I know off.

So was wondering if you had thought/looked at some of this and
had ideas  - and were thinking to do something in the future?

Thanks!

P.S.
I hope I have CC-ed everybody here.