On Mon, 7 Feb 2011 19:50:40 +0000 Simon Wilkinson <[email protected]> wrote:
> On 7 Feb 2011, at 19:42, Derrick Brashear wrote: > > > I believe it to be possible but unlikely. > > It's also not possible to fix it just by changing the client behaviour > to only write dirty bytes (rather than dirty chunks) to the server. On > Linux, for example, we'll always write and read full pages. So, if you > were to write bytes 16-31, the kernel would ask AFS to read bytes > 0->4095 into memory, before performing the write. If you don't have > that page locally, then you would lose. Well, shoot. Scratch that, then. But even then, I agree that that alone is not a reason not to do this. Just something to make people aware of for whatever release (or something that is configurable, if someone cares enough about it, but I'd expect not). -- Andrew Deason [email protected] _______________________________________________ OpenAFS-devel mailing list [email protected] https://lists.openafs.org/mailman/listinfo/openafs-devel
