Overview

SpeedLink is a high throughput, low latency compute paradigm implemented using GridServer. SpeedLink can deliver end-to-end compute latencies of less than one millisecond, depending on the server and network environment, with no theoretical upper limit on aggregate grid-wide operation throughput. Individual clients can encounter a practical throughput limit based on the client system's performance.

SpeedLink uses the Engine pool of a GridServer installation, sharing Engines seamlessly with other GridServer Services. SpeedLink maintains key GridServer features like Service prioritization and Engine balancing, and standard Services can co-exist with SpeedLink Services, even within the same client or Engine process. SpeedLink Services request Engines from a GridServer Broker, and this allocation operates much like a standard GridServer Service invocation. Unlike standard Services, allocated Engines communicate directly with the SpeedLink client, resulting in the desired throughput and latency characteristics. Allocated Engines are periodically returned to the Engine pool.

SpeedLink's design supports easy code migration from standard GridServer Services to SpeedLink Services, and back again to standard Services if desired. The SpeedLink API provides the Service interface for SpeedLink Services. The application programmer simply calls the createService method on a different factory class. The resulting Service interface (or virtual object, depending on the chosen implementation language) can be used the same way you use a standard Service.