Modelling and programming stream-based distributed computing based on the meta-pipeline approach. ... The Caravela platform uses GPUs as processing units to execute multiple flow-models, connected by means of input-output streams, using a pipeline-model. The paper describes the basic theoretical background to the platform, as well as the meta-pipeline mechanism implemented by using it. Caravela platform establishes a distributed parallel computing environment through a virtual network using dedicated servers: broker servers act as routers to reach worker servers, i.e. machines hosting a video adapter with a single or multiple GPUs, i.e. shaders. A flow-model, consisting of input-output streams, constant parameters and program, can be executed in a local machine or in remote machines connected through the network. The Caravela library provides functions for executing a flow-model mapped to a shader, getting a buffer of input or output data stream, etc. The paper focuses on an extension of the pipelined processing mechanism, directly connecting flow-models mapped to multiple shaders distributed across the Caravela platform.
Keywords for this software
References in zbMATH (referenced in 2 articles )
Showing results 1 to 2 of 2.
- Falcão, Gabriel; Yamagiwa, Shinichi; Silva, Vitor; Sousa, Leonel: Parallel LDPC decoding on GPUs using a stream-based computing approach (2009)
- Yamagiwa, Shinichi; Sousa, Leonel: Modelling and programming stream-based distributed computing based on the meta-pipeline approach (2009)