Accelerator Offloading Allocation for Balanced Server Throughput
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The capacitance limitations of electronic circuitry in FPGAs make it difficult to offload multiple types of APLs in a server cluster, leading to unbalanced allocation and decreased overall throughput.
Innovation Solution
A server control apparatus that optimizes the allocation of accelerators by determining a performance-to-request ratio and adjusting the allocation to ensure variance within a predetermined threshold, using a network system with a virtual patch panel and switch to dynamically manage connections between applications and accelerators.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If accelerators are allocated to multiple APLs in an FPGA cluster, then the processing capability and throughput should improve, but the limited capacitance of electronic circuitry in FPGAs causes unbalanced allocation and prevents effective offloading of multiple APL types
Solution Approach 1:
The patent implements dynamic allocation of accelerators to APLs based on real-time workload conditions. The server control apparatus monitors the state of each APL and accelerator, and adjusts the allocation relationships dynamically. This allows the system to adapt to changing demands and optimize throughput by ensuring that accelerators are allocated to the APLs that need them most at any given time, resolving the contradiction between improving productivity through multi-APL offloading and the limited capacitance constraints of FPGAs.
2Productivity
If accelerators are allocated to specific APLs, then processing performance improves, but unbalanced allocation leads to overload or underutilization, decreasing overall system throughput
Solution Approach 1:
The patent employs a feedback mechanism where the server control apparatus continuously monitors the workload and performance metrics of each APL-accelerator pair. Based on this feedback information, the control apparatus adjusts the allocation of accelerators to maintain balanced utilization. When an accelerator becomes overloaded, the system reallocates its workload to other available accelerators, preventing system bottlenecks and ensuring reliable, balanced operation across the cluster.
3Productivity
If manual optimization of accelerator allocation is performed, then some APLs can be offloaded effectively, but the complexity of managing multiple APL types with limited FPGA capacitance increases significantly
Solution Approach 1:
The patent implements a self-service mechanism where the server control apparatus automatically manages accelerator allocation without requiring manual intervention. The control apparatus monitors the state of APLs and accelerators, determines optimal allocation configurations, and executes the allocation adjustments autonomously. This self-service approach maintains high offloading efficiency while eliminating the complexity of manual management, as the system handles the optimization automatically based on real-time conditions.
Data Source
AI summary
A server control apparatus 100 includes a request and configuration collection unit 120 that acquires a request to offload a certain process of an application to an accelerator for each application in a server, and configurations of the accelerator and the application in a server 30, an optimization arithmetic unit 130 that determines, by referring to information of the acquired request and configurations of the server 30, a ratio of processing performance to the request, and optimizes allocation of the accelerator so that variance of the ratio between the applications is equal to or less than a predetermined threshold, and a configuration determination unit 140 that determines a configuration suggestion to be taken by the server 30 by using an arithmetic result from the optimization arithmetic unit 130 and a predetermined policy, and commands the server 30 to execute the configuration suggestion.