[PVFS2-developers] Mechanism for forwarding data
rross at mcs.anl.gov
Mon Jul 12 20:42:04 EDT 2004
I agree with Neill, although I have thought some about using Trove for
shared access to storage, so if the flow option turns out to look really
nasty, we might want to revisit.
I think if we were going to do some sort of Trove option, it might look
more like a pair of writes to Trove, or a write that goes to two different
file handles (one of them residing locally, the other remotely).
One disadvantage of Trove in that mode is that it only takes offset-length
pairs right now, so noncontiguous I/O could be ugly.
Anyway, I'd go with the flow approach. The error cases are going to be
On Wed, 7 Jul 2004 neillm at mcs.anl.gov wrote:
> On Wed, Jul 07, 2004 at 01:53:22PM -0400, Bradley W Settlemyer wrote:
> > Is either one of these approaches superior to the other? Is there
> > another simple or simpler approach that is superior to these?
> A new flow protocol (that can be config enabled like the others) is my
> vote. Starting a flow from trove is not something I'd like to see ;-)
More information about the PVFS2-developers