[Spread-users] Spread questions
mike at marketocracy.com
Mon Aug 12 01:09:14 EDT 2002
I'm new to Spread and the entire 'Virtual Synchrony' model that it is based
on. I'm trying to build a replicated key/value data store. In the VS
model, am I still required to hold coordinator elections and do a two-phase
commit protocol in order to ensure that the data stays consistent , or will
VS take care of these details for me as part of the model? I'm working on
building a prototype that ends up running 3 parallel state machines
(synchronization, coordinator election, and two-phase commit), and it's
starting to get hairy. I'm just wondering if I'm over-thinking this problem
and don't really need to be doing all of this work.
My general goals are:
- Clients can connect to any node in the cluster.
- Reads can come from only that node.
- Writes atomically get committed to all nodes.
- When a new node joins, it 'bootstraps' it's state from
pre-existing nodes before becoming available for client use.
More information about the Spread-users