CTDB - the 'Mainz' plan for clustered Samba
tridge at au1.ibm.com
Tue Oct 3 11:11:52 GMT 2006
> I drew first diagrams about Samba & Clustering and placed them at the
> wiki page http://wiki.samba.org/index.php/Samba_%26_Clustering. Please,
> check if I didn't understand your ideas.
A couple of minor things:
- some of the steps listed in the diagrams will be skipped in the
average case. I should explain more about the way that nodes will
cache DMASTER lookups. Basically a node can agressively cache any
DMASTER lookup it does. It doesn't matter if the cache goes stale,
as a old cache entry will trigger an error which will cause a new
lookup of the DMASTER from the LMASTER.
- I expect that the 'conditional append' and 'remove trigger'
operations will be the most common operations, not the 'fetch'
operation. It basically allows us to combine a
lock/fetch/modify/write/unlock sequence into a single round
trip. This builds upon Volkers fetch_and_lock method which reduced
it to two round trips.
More information about the samba-technical