Graph DASL:a Domain specific language for graphs with Accelerated Scala using Linear algebra

Representing networks and complex associations of all types, graphs are a critically important data structure within the Department of Defense.Applying graph analytics and mining to large quantities of graph data enables conducting operations, gathering intelligence, and defending the nations critical capabilities.Existing capabilities for graph mining are limited to modest data scales, batch processing, and have iterative algorithms that are complex to express in existing graph pattern languages or in data-parallel graph constructs.These algorithms require a procedural (vs declarative) definition.They form an important class of techniques that enable community detection, deep-learning, and other techniques.Apache Spark is a rapidly emerging platform for complex analytics, which uses the Scala functional language.GPUs have been demonstrated to provide 1000X acceleration graph and linear algebra problems. DASL is a Scala based DSL plugin for GPU accelerated graph analytics on Spark.DASL will be 100-1000x faster than Spark, 10,000x faster than Hadoop, and scale from laptops to clusters.DASL will reduce the barrier of use for GPUs on graphs, enable simpler and smarter algorithms, and deliver the performance of GPUs through a library of fast linear algebra primitives and scalable runtime.We will demonstrate DASL on Cyber domain problems.