Scopes

The scope of a vertex is the data on the vertex, the data on all adjacent edges as well as data on all adjacent vertices. An update function executed on a vertex has the ability to read and write to all data within the scope of the vertex.

The ability for an update function to read or write a piece of data does not mean that it can do so in a consistent fashion. For instance, without adequate protection, it is conceivable that update functions executed on neighboring vertices could modify the same piece of data at the same time.

GraphLab therefore provides the concept of a consistency class which expresses both the level of protection provided, as well as the amount of parallelism available. The consistency type can be selected through the graphlab::iengine::set_default_scope() function in the graphlab::iengine, or the ref graphlab::core::set_scope_type() function in the graphlab::core .

There are three basic types of scopes (as well a number of undocumented ones which you probably should not use). The scopes are currently implemented through the use of a read-write lock on each vertex. Different types of scopes then make use of different locking schemes.

• Full Consistency This scope type is the safest and guarantees safe access to the current vertex, adjacent edges, and adjacent vertices. This is implemented by acquiring a write lock on all vertices in the scope.
• Edge Consistency This scope type guarantees safe read/write access to the current vertex as well as adjacent edges. In addition, you can also read (but not write) data on adjacent vertices safely. This is implemented by acquiring a write lock on the current vertex, and a read lock on all adjacent vertices.
• Vertex Consistency This scope type guarantees only safe read/write access to the current vertex. This is implemented by acquiring a write lock on the current vertex.