Earticle

최근 GPU가 그래픽 처리뿐 아니라 다양한 분야의 병렬 처리로 그 영역을 넓혀가고 있다. 그러나, 현재 GPU는 워크로드의 다양성을 반영하기보다 간결한 제어 구조를 통한 개별 워크로드의 병렬성 극대화에 초점을 맞추고 있다. 본 논문은 워크로드 특성을 반영한 GPU 작업 배치를 위해 GPU에서 수행되는 워크로드의 자원 사용 특성을 컴퓨팅 바운드형, 메모리 바운드형, 실행종속 지연형으로 분류한 후, 각 분류에서 병목점이 되는 세부 자원을 규명한다. 예를 들어 컴퓨팅 바운드형의 경우 단정밀도 연산장치, 배정밀도 연산장치, 특수함수 연산장치 등 병목 자원이 무엇인지 분석한다. 본 논문의 분석 결과는 동일한 컴퓨팅 바운드형 워크로드라도 병목이 되는 세부 자원이 다를 경우 함께 배치하는 것이 성능 충돌을 일으키지 않는다는 점을 규명하여 GPU 작업배치의 효율화에 기여할 것으로 기대된다.

Recently, GPU expands application domains from graphic processing to various kinds of parallel workloads. However, current GPU systems focus on the maximization of each workload’s parallelism through simplified control rather than considering various workload characteristics. This paper classifies the resource usage characteristics of GPU workloads into computing-bound, memory-bound, and dependency-latency-bound, and quantifies the fine-grained bottleneck for efficient workload allocation. For example, we identify the exact bottleneck resources such as single function unit, double function unit, or special function unit even for the same computing-bound workloads. Our analysis implies that workloads can be allocated together if fine-grained bottleneck resources are different even for the same computing-bound workloads, which can eventually contribute to efficient workload allocation in GPU.

The current and future computer industry is changing from single core CPU to Multicore Processors. The next generation computing systems are focusing on parallel computing to solve the problem in fast and high speed using parallel programming concepts. By running more than one task at the same time with multiple processors concurrently or parallel we can achieve high speed in our computing applications. Here the performance improvement is measured in terms of increase in the number of cores per machine and is analyzed for better optimal work load balance. The parallel computers follow different workload scheduler. In this paper we investigate how to tune the performance of threaded applications with balanced load for each core or processor. The focus here is the process of comparative analysis of single core and multicore systems to run an application program for faster execution time and optimize the scheduler for better performance.