-
Notifications
You must be signed in to change notification settings - Fork 7
Distribution
uwiger edited this page Jul 5, 2012
·
1 revision
There are a number of issues to nail down for the distribution support:
- Scalability: how do we want to scale? Smart distribution of device groups and users, as well as load-balancing feeds, will probably inform the data model scalability
- Consistency: Replication and netsplit tolerance; kvdb is prepared for automatic attributes (e.g. vector clocks).
- Locking: need to avoid deadlocks and livelocks, while not killing scalability or jeopardizing consistency
TBD
One option is to grab one of the existing frameworks, e.g.
- Riak core (although Dizzy doesn't recommend this, considering our requirements)
- HyperDex (but haven't really looked into this yet)
I (Ulf) am finalizing a slightly unusual locking library, called ddd(Distributed Deadlock Detection). The underlying algorithm was model-checked by Thomas Arts several years ago (I came up with it in 1992), but it has never been 'productified'. See more on Locking.