[pnfs] Parallel tasks

[Iyer, Rahul]

Hi,
We had briefly discussed the tasks that could be donbe in parallel
during the last pNFS call. I apologize for taking so long to pen this
down. The way I see it, here are some of the tasks that can go on in
parallel:
 
1. Client side layout cache implementation and adapting the current code
to use the cache instead of inode->current_layout. Andy, do you have
some of this code from your blocks client?
 
2. Server side layout cache implementation
 
As a simplifying assumption, for the first cut, we could still use only
whole file layouts. This would mean that the layout cache would have
only one layout segment entry per layout. This is similar to our current
client code, except it changes where the layout is found. Not entirely
sure what this means on the server, but I think this would require no
server side changes. Later on we could add segments and then the server
would need to handle segment management.
 
3. Session-ifying the recalls. The current recall code has no sessions
stuff in it, like CB_SEQUENCE and the like. We currently have slot table
implementations for the client and the server. We could use this for the
callbacks too. We leave the rest unchanged; i.e, we make a separate RPC
connection to the client and issue callbacks (CB_RECALL, LAYOUTRECALL
etc.) When my code is done, we can integrate these two codelines and
we'll have a correct callback implementation.
 
And finally, in case 1, 2 and 3 are done, then we could go about getting
the LAYOUTRECALL code to use the cache and make sure it works.
 
All these activities are independent of my transport level callback
implementation and can go on in parallel with my development.
 
Thanks
Regards
Rahul
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://linux-nfs.org/pipermail/pnfs/attachments/20070416/ca20b2fa/attachment.htm