The creators of GraphLab use two different types of engines to implement the asynchronous execution model. The first one is called the Chromatic Engine, and it uses vertex coloring to color vertices of a graph so that no two adjacent nodes share the same color. Because nodes of the same color are nonadjacent and can be processed in parallel, the Chromatic fulfills Edge Consistency (see slide 9).
The other engine is called the Locking Engine. Whereas the Chromatic Engine executes some nodes in parallel, the Locking Engine fully executes all nodes asynchronously via a distributed locking system. In particular, this engine utilizes pipelined locking, allowing each machine to request locks for many scopes at once in an effort to hide latency.
For more information about CMU's GraphLab, you can check out the authors' work here.
The creators of GraphLab use two different types of engines to implement the asynchronous execution model. The first one is called the Chromatic Engine, and it uses vertex coloring to color vertices of a graph so that no two adjacent nodes share the same color. Because nodes of the same color are nonadjacent and can be processed in parallel, the Chromatic fulfills Edge Consistency (see slide 9).
The other engine is called the Locking Engine. Whereas the Chromatic Engine executes some nodes in parallel, the Locking Engine fully executes all nodes asynchronously via a distributed locking system. In particular, this engine utilizes pipelined locking, allowing each machine to request locks for many scopes at once in an effort to hide latency.
For more information about CMU's GraphLab, you can check out the authors' work here.
This comment was marked helpful 0 times.