Matrix engines or units, in different forms and affinities, are becoming a reality in modern processors; CPUs and otherwise. The current and dominant algorithmic approach to Deep Learning merits the commercial investments in these units, and deduced from the No. 1 benchmark in supercomputing, namely High Performance Linpack, one would expect an awakened enthusiasm by the HPC community, too. Hence, our goal is to identify the practical added benefits for HPC and machine learning applications by having access to matrix engines. For this purpose, we perform an in-depth survey of software stacks, proxy applications and benchmarks, and historical batch job records. We provide a cost-benefit analysis of matrix engines, both asymptotically and in conjunction with state-of-the-art processors. While our empirical data will temper the enthusiasm, we also outline opportunities to “misuse” these dense matrix-multiplication engines if they come for free.
Modern supercomputer performance is principally limited by power. TSUBAME-KFC is a state-of-the-art prototype for our next-generation TSUBAME3.0 supercomputer and towards future exascale. In collaboration with Green Revolution Cooling and others, TSUBAME-KFC submerges compute nodes configured with extremely high processor/component density, into non-toxic, low viscosity oil with high 260 Celsius flash point, and cooled using ambient / evaporative cooling tower. This minimizes cooling power while all semiconductor components kept at low temperature to lower leakage current. Numerous off-line in addition to on-line power and temperature sensors are facilitated throughout and constantly monitored to immediately observe the effect of voltage/frequency control. As a result, TSUBAME-KFC achieved world No.1 on the Green500 in Nov. 2013 and Jun. 2014, by over 20% c.f. the nearest competitors.
Performance prediction of parallel applications across systems becomes increasingly important in today's diverse computing environments. A wide range of choices in execution platforms pose new challenges to researchers in choosing a system which best fits their workloads and administrators in scheduling applications to the best performing systems. While previous studies have employed simulation-or profile-based prediction approaches, such solutions are time-consuming to be deployed on multiple platforms. To address this problem, we use two collaborative filtering techniques to build analytical models which can quickly and accurately predict the performance of workloads across different multicore systems. The first technique leverages information gained from performance observed for certain applications on a subset of systems and use it to discover similarities among applications as well as systems. The second collaborative filtering based model learns latent features of systems and workloads automatically and use these features to characterize the performance of applications on different platforms. We evaluated both the methods using 30 workloads chosen from NAS Parallel Benchmarks, BOTS and Rodinia benchmarking suites on ten different systems. Our results show that such collaborative filtering methods can make predictions with RMSE as low as 0.6 and with an average RMSE of 1.6.
1 April 2016 in Electronic Proceedings in Theoretical Computer Science
In this paper, we prove a version of the typed B\"ohm theorem on the linear lambda calculus, which says, for any given types A and B, when two different closed terms s1 and s2 of A and any closed terms u1 and u2 of B are given, there is a term t such that t s1 is convertible to u1 and t s2 is convertible to u2. Several years ago, a weaker version of this theorem was proved, but the stronger version was open. As a corollary of this theorem, we prove that if A has two different closed terms s1 and s2, then A is functionally complete with regard to s1 and s2. So far, it was only known that a few types are functionally complete.Comment: In Proceedings MSFP 2016, arXiv:1604.0038
Existing data grid scheduling systems handle huge data I/O via replica location services coupled with simple staging, decoupled from scheduling of computing tasks. However, when the application/workflow scales, we observe considerable degradations in performance, compared to processing within a tightly-coupled cluster. For example, when numerous nodes access the same set of files simultaneously, major performance degradation occurs even if replicas are used, due to bottlenecks that manifest in the infrastructure. Instead of resorting to expensive solutions such as parallel file systems, we propose alleviating the situation by tightly coupling replica and data transfer management with computation scheduling. In particular we propose three techniques: (1) dynamic aggregation and O(1) replication of data-staging requests across multiple nodes using a multi-replication framework, (2) replica-centric scheduling - data re-use and time-to-replication as compute scheduling metrics on the grid and (3) overlapped execution of data staging and compute bound tasks. Early benchmark results implemented in our prototype Condor-like grid scheduling system demonstrate that the techniques are quite effective in eliminating much of the overhead in data transfers in many cases