The core innovation of ThroughPuter PaaS, the Dynamic Parallel Execution Environment (DPEE), is designed to support executing multiple concurrent (internally pipelined and/or parallelized) application programs with highly time-variable processing loads, under app-specific latency constraints, on dynamically shared manycore processor hardware, with maximized resource-efficiency and on-time throughput performance.
To achieve this fundamental capability needed for scalable, high-performance and cost-efficient cloud computing in the era of dynamic inter- and intra-application parallel processing, the execution hardware of ThroughPuter PaaS had to rise up to the challenge: instead of merely providing multiple processing cores to process multiple instruction streams in parallel, the intelligent manycore fabric of the DPEE automates the runtime platform operating system functions in the hardware, thus preserving the processor cores for executing dynamically optimized sets of the user application tasks.
A key realization leading to implementing the platform OS functions in hardware is that, if system software was relied on for the time-critical dynamic resource management functions (such as optimizing allocations of processing resources among the multiple apps, assignments of the processing cores to the appropriate tasks of the app instances, connecting inter-task and IO communications data, updating fast-access memory contents, security policing and billing for the apps, load balancing for the incoming requests etc.), the system software overhead would quickly clog up the processing hardware -- and particularly so with the workloads becoming dominated by concurrent processing of short-lived and/or latency-sensitive transactions, and only the more so when seeking to theoretically increase scalability through sharing larger pools of processing resources among greater numbers of concurrent apps.
To resolve this severe scalability constraint caused by system software overhead, in the design of the execution environment of ThroughPuter PaaS, it was understood that for truly scalable cloud processors, it is just as essential for the hardware to implement the runtime OS functionality (incl. dynamic resource management per above, to achieve optimized use of the manycore processing capacity) as it is to execute the actual client application program instructions.
The resulting architectural performance, cost-efficiency and inter-app security advantages of the DPEE of ThroughPuter PaaS are made available, with high development productivity, for the cloud app developers and service providers through the integrated parallel program development environment of the PaaS. As a result, ThroughPuter platform architecture is cross-layer optimized for dynamic, high-performance, high-efficiency, secure cloud computing. Collectively, the dynamic parallel processing platform achieves maximized processing on-time throughput, per unit cost, across all client programs dynamically and securely sharing the pooled processing resources.
For illustration, please review technical overview of ThroughPuter PaaS solution:
http://www.throughputer.com/uploads/ThroughPuterPaaSforParallelProcessing.pdf
Quick Features -> Benefits -> Customer Value sheet:
http://www.throughputer.com/uploads/DynamicParallelExecution-FeaturesBenefitsValue.pdf
A short paper of ThroughPuter's key innovations enabling scalable high performance cloud computing: Hardware Implemented Runtime System For Dynamically Shared Manycore Processors:
http://www.throughputer.com/uploads/HardwareOSforDynamicallySharedManycoreProcessors.pdf
A paper on the economic model for ThroughPuter PaaS contracts:
For full technical reference on ThroughPuter PaaS and its Dynamic Parallel Execution Environment, please contact us via tech@throughputer.com.