Grid Computing Edition

Starting with version 4.0, ScaleOut Software introduces ScaleOut StateServer Grid Computing Edition™ containing powerful new features that make distributed caching a vital component of grid computing environments. Grid Computing Edition combines ScaleOut StateServer's highly scalable data caching with exciting new capabilities for rapidly searching cached data and quickly developing scalable, grid-based applications.

Distributed Data Caching

Grid computing has rapidly grown in popularity to address the needs of complex computational problems which require scalable performance that only compute grids can provide. Financial services and other data-intensive industries routinely demand real time processing and overnight batch analysis of large data sets. By storing fast-changing data in ScaleOut StateServer's distributed, in-memory cache, these applications can dramatically reduce access latencies, avoid bottlenecks, and achieve peak performance. Distributed caching also lets applications immediately share data across compute nodes without the need for message passing; this simplifies program structure and shortens design cycles. These combined benefits make distributed caching a powerful data access platform for a wide range of HPC applications.

Powerful Parallel Cache Search

ScaleOut StateServer Grid Computing Edition further extends data caching's potential to maximize application performance and simplify design. New grid computing capabilities let applications easily access and operate in parallel on stored data across the compute grid. Applications can now perform parallel queries to rapidly search the distributed cache for selected objects based on metadata associated with cached objects. For example, cached stock price objects could be searched to determine which stocks were associated with a specified industry group. Employing patent-pending, parallel search and merging algorithms, ScaleOut StateServer provides the fastest possible parallel query across all hosts within the distributed cache.

Parallel Method Invocation

ScaleOut StateServer Grid Computing Edition takes data grid support to the next level by enabling applications to execute user-defined methods in parallel on a selected set of objects and then combine the results using user-defined merge algorithms. For example, a set of portfolio objects could be analyzed in parallel with the results merged into a single report. By simplifying application design and reducing data motion, this "map-reduce" capability works in concert with the grid's job scheduler to further extend distributed caching's power to accelerate the performance of HPC applications.

Unlike traditional map/reduce implementations which process large, file-based data sets, Grid Computing Edition stores data sets in ScaleOut StateServer's scalable, distributed cache as a large set of cached objects. Grid Computing Edition automatically maps the user's "map" operation across both multiple cores within each server and multiple servers within the grid, and the map operation then executes in parallel on a queried set of cached objects. GCE's runtime system maximizes parallelism and minimizes data motion within the grid, In addition, GCE performs high-speed, parallel execution of the user's "reduce" operation across all grid nodes. The result is very high performance for the complete map/reduce operation.

No Need for Special Code

Importantly, the user does not need to write special code to harness the power of the grid's infrastructure. Grid Computing Edition's APIs enable map/reduce calculations to be written as in-memory methods that avoid explicit cache accesses, and it provides automatic parallel speedup. In addition, the need for traditional HPC message-passing within the user's application is eliminated. Now, developers can obtain parallel performance without becoming a parallel processing expert.