26 September 2016 in IEEE Transactions on Parallel and Distributed Systems
The non-deterministic nature of multi-threaded workloads running on multi-core platforms often leads to notable performance variability from run to run. Such variability makes experimental results prone to misinterpretations or misguided claims. To deal with such variability, statistical inference methods are usually used to summarize the experimental results with certain confidence levels by running the experiments or measurements a large number of times. However, such statistical results are often too vague or too simplistic. They are not sufficient to help users understand the causes of such variability, and allow more in-depth analysis on the results or reproduce the results for validation during design space exploration. To allow better analyzability and reproducibility, we propose a framework to tackle such variability, called VarCatcher. The key to VarCatcher is to characterize a parallel execution using Parallel Characteristics Vector (PCV). A clustering-based approach is then used to group runs with similar execution characteristics that can later be used to analyze results in-depth, to customize different evaluation strategies, reproduce the result for variability, to determine the impact of features, or to assist performance diagnosis. We have built a prototype of VarCatcher that includes a user-level toolset for runtime monitoring and measurements using the Intel Processor Trace feature on commodity Intel processors as well as an architecture extension with very low runtime overheads (around 3 and 0.01 percent accordingly). Several case studies confirm that VarCatcher enables several appealing features such as in-depth result analysis, customized evaluation strategies, and reproducibility.
24 April 2023 in IEEE Transactions on Parallel and Distributed Systems
Graph analytics are widely used including recommender systems, scientific computing, and data mining. Meanwhile, GPU has become the major accelerator for such applications. However, the graph size increases rapidly and often exceeds the GPU memory, incurring severe performance degradation due to frequent data transfers between the main memory and GPUs. To relieve this problem, we focus on the utilization of data in GPUs by taking advantage of the data reuse across iterations. In our studies, we deeply analyze the memory access patterns of graph applications at different granularities. We have found that the memory footprint is accessed with a roughly sequential scan without a hotspot, which infers an extremely long reuse distance. Based on our observation, we propose a novel framework, called Liberator , to exploit the data reuse within GPU memory. In Liberator , GPU memory is reserved for the data potentially accessed across iterations to avoid excessive data transfer between the main memory and GPUs. For the data not existing in GPU memory, a Merged and Aligned memory access manner is employed to improve the transmission efficiency. We also further optimize the framework by parallel processing of data in GPU memory and data in the main memory. We have implemented a prototype of the Liberator framework and conducted a series of experiments on performance evaluation. The experimental results show that Liberator can significantly reduce the data transfer overhead, which achieves an average of 2.7x speedup over a state-of-the-art approach.
1 August 1992 in ACM Transactions on Computer Systems
In large multiprocessor systems, fast synchronization is crucial for high performance. However, synchronization traffic tends to create “hot-spots” in shared memory and cause network congestion. Multistage shuffle-exchange networks have been proposed and built to handle synchronization traffic. Software combining schemes have also been proposed to relieve network congestion caused by hot-spots. However, multistage combining networks could be very expensive and software combining could be very slow. In this paper, we propose a single-stage combining network to handle synchronization traffic, which is separated from the regular memory traffic. A single-stage combining network has several advantages: (1) it is attractive from an implementation perspective because only one stage is needed(instead of log N stages); (2) Only one network is needed to handle both forward and returning requests; (3) combined requests are distributed evenly through the network—the wait buffer size is reduced; and (4) fast-finishing algorithms [30] can be used to shorten the network delay. Because of all these advantages, we show that a single-stage combining network gives good performance at a lower cost than a multistage combining network.
2 May 2017 in IEEE Transactions on Parallel and Distributed Systems
Most existing computer architecture simulators are cycle oriented, i.e., they are driven cycle by cycle. However, frequent switches among simulation contexts, excessive buffer accesses and tightly coupled manner often make such an architecture simulator slow, difficult to parallelize and hard to scale to large-scale many-core systems. In this paper, we propose Prophet, a parallel instruction-oriented simulation framework for many-cores. Prophet adopts a general instruction-oriented model to simulate processor cores, in which a simulator is built from the perspective of each simulated instruction impacting a small number of relevant processor components, as opposed to that of a large number of processor components executing many instructions in each cycle as in the cycle-oriented approach. Prophet determines the execution cycle of a simulated instruction based on the states of the relevant components impacted by the instruction, and update the components states after the execution of the instruction. Prophet also adopts a speculative model to decouple private resources from the shared resources (e.g., shared cache), which avoids unnecessary interactions between them and only pays a penalty upon a rare mis-speculation. We have designed and implemented a prototype of Prophet that supports both user-level and full-system simulation. Experimental results show Prophet can scale up to simulate thousands of simulated cores (4,096 cores in the current implementation) with good performance and small accuracy loss. It achieves average simulation speeds of about 98 and 235 MIPS (millions of simulated instructions per second) for full-system and user-level simulation, respectively, with only 3 percent IPC error rate and negligible deviation in cache simulation results. When run on a many-core platform (i.e., Intel Xeon Phi), it achieved an average simulation speed of about 413 MIPS.