System and micro-architectural level analysis of high throughput workloads. High Throughput Computers project, Institute of Computing Technology, Chinese Academy of Sciences, Beijing.Moreover, the generated proxy benchmarks reflect consistent performance trends across different architectures. The proxy benchmarks shorten the execution time by 100s times on real systems while maintaining the average system and micro-architecture performance data accuracy above 90%, even changing the input data sets or cluster configurations. We propose a data motif-based proxy benchmark generating methodology which combines data motifs with different weights to mimic the big data and AI workloads. Proxy Benchmark Generating for Big Data and AI. The latest version of BigDataBench is available from. BigDataBench 4.0 adopts a scalable data motif based benchmarking methodology and contains 13 representative real-world data sets and 47 benchmarks. Significantly different from the traditional kernels, a data motif’s behaviors are affected by the sizes, patterns, types, and sources of different data inputs Moreover, it reflects not only computation patterns, memory access patterns, but also disk and network I/O patterns. We consider each big data and AI workload as a pipeline of one or more classes of unit of computation performed on different initial or intermediate data inputs, each of which captures the common requirements while being reasonably divorced from individual implementations. Data motif is a new approach to modelling and characterizing big data and AI workloads. Straightforwardly, we can tailor the architecture to characteristics of an application, several applications, or even a domain of applications. Identifying abstractions of time-consuming units of computation is an important step toward domain-specific hardware and software co-design. Data Motifs: A Lens towards Fully Understanding Big Data and AI Workloads.We abstract the HPC AI system into nine independent layers, and present explicit benchmarking rules and procedures to assure equivalence of each layer, repeatability, and replicability. HPC AI500 presents a comprehensive methodology, tools, Roofline performance models, and innovative metrics for benchmarking, optimizing, and ranking HPC AI systems. HPC AI500: A Benchmark Suite for HPC AI Systems.We present a comprehensive AI inference benchmark suite including nineteen real-world AI workloads and covering text processing, image processing, audio processing, video processing, and 3D data processing. AIBench Inference: Comprehensive AI Inference Benchmarks.We give the hotspot functions as microbenchmarks (AIBench Micro) to achieve portability for simulator-based research after profiling. Besides, we propose two AIBench Training subset: RPR and WC subsets to achieve affordability. We provide nineteen real-world AI workloads to achieve comprehensiveness and representativeness. Our methodology widely investigates AI tasks and models and covers the algorithm-level, system-level, and microarchitectural-level factors space to the most considerable extent. We present a balanced methodology considering comprehensiveness, representativeness, affordability, and portability. AIBench Training: Balanced AI Training Benchmarks.Each scenario benchmark models the critical paths of a real-world application scenario as a permutation of the AI and non-AI modules. AIBench Scenario is a benchmarks suite modeling AI-augmented Internet service scenarios, including E-commerce Search Intelligence and Online Translation Intelligence. Based on the framework, we implement the scenario benchmarks-AIBench Scenario, including E-commerce Search Intelligence and Online Translation Intelligence. We design and implement a reusing benchmark framework. We propose a scenario-distilling methodology that formalizes a real-world or future application scenario as a Directed Acyclic Graph-based model (in short, DAG model). ScenarioBench: Modeling, Characterizing, and Optimizing Ultra-scale Real-world or Future Applications and Systems Using Benchmarks.My research interests focus on the following points of Data center computing and benchmarking:
0 Comments
Leave a Reply. |