Task-parallel Applications Research Articles

Experiences with nested parallelism in task-parallel applications using malleable BLAS on multicore processors

Malleability is defined as the ability to vary the degree of parallelism at runtime, and is regarded as a means to improve core occupation on state-of-the-art multicore processors tshat contain tens of computational cores per socket. This property is especially interesting for applications consisting of irregular workloads and/or divergent executions paths. The integration of malleability in high-performance instances of the Basic Linear Algebra Subprograms (BLAS) is currently nonexistent, and, in consequence, applications relying on these computational kernels cannot benefit from this capability. In response to this scenario, in this paper we demonstrate that significant performance benefits can be gathered via the exploitation of malleability in a framework designed to implement portable and high-performance BLAS-like operations. For this purpose, we integrate malleability within the BLIS library, and provide an experimental evaluation of the result on three different practical use cases.

Task-RM: A Resource Manager for Energy Reduction in Task-Parallel Applications under Quality of Service Constraints

Improving energy efficiency is an important goal of computer system design. This article focuses on a general model of task-parallel applications under quality-of-service requirements on the completion time. Our technique, called Task-RM , exploits the variance in task execution-times and imbalance between tasks to allocate just enough resources in terms of voltage-frequency and core-allocation so that the application completes before the deadline. Moreover, we provide a solution that can harness additional energy savings with the availability of additional processors. We observe that, for the proposed run-time resource manager to allocate resources, it requires specification of the soft deadlines to the tasks. This is accomplished by analyzing the energy-saving scenarios offline and by providing Task-RM with the performance requirements of the tasks. The evaluation shows an energy saving of 33% compared to race-to-idle and 22% compared to dynamic slack allocation (DSA) with an overhead of less than 1%.

Design and Implementation of a Criticality- and Heterogeneity-Aware Runtime System for Task-Parallel Applications

Heterogeneous multiprocessing (HMP) is an emerging technology for high-performance and energy-efficient computing. While task parallelism is widely used in various computing domains, such as embedded, big-data, and machine-learning computing domains, it still remains unexplored to investigate the efficient runtime support that effectively utilizes the criticality of the tasks of the target application and the heterogeneity of the underlying HMP system with full resource management. To bridge this gap, we propose CHRT, a criticality- and heterogeneity-aware runtime system for task-parallel applications. CHRT dynamically estimates the performance and power consumption of the target task-parallel application and robustly manages the full HMP system resources (i.e., core types, counts, and voltage/frequency levels) to maximize the overall efficiency. Our quantitative evaluation based on widely-used task parallel benchmarks and two full HMP systems (i.e., the XU3 and HiKey970 HMP systems) demonstrates the effectiveness of CHRT in that CHRT achieves significantly higher energy (e.g., 60.4 and 57.2 percent on average on the XU3 system) and energy-delay product (e.g., 52.2 and 44.0 percent on average on the HiKey970 system) efficiency than the baseline runtime system that employs the breadth-first scheduler and the state-of-the-art criticality-aware runtime system and incurs low performance overheads.

Analysis and Optimization of Task Granularity on the Java Virtual Machine

Task granularity, i.e., the amount of work performed by parallel tasks, is a key performance attribute of parallel applications. On the one hand, fine-grained tasks (i.e., small tasks carrying out few computations) may introduce considerable parallelization overheads. On the other hand, coarse-grained tasks (i.e., large tasks performing substantial computations) may not fully utilize the available CPU cores, leading to missed parallelization opportunities. In this article, we provide a better understanding of task granularity for task-parallel applications running on a single Java Virtual Machine in a shared-memory multicore. We present a new methodology to accurately and efficiently collect the granularity of each executed task, implemented in a novel profiler (available open-source) that collects carefully selected metrics from the whole system stack with low overhead, and helps developers locate performance and scalability problems. We analyze task granularity in the DaCapo, ScalaBench, and Spark Perf benchmark suites, revealing inefficiencies related to fine-grained and coarse-grained tasks in several applications. We demonstrate that the collected task-granularity profiles are actionable by optimizing task granularity in several applications, achieving speedups up to a factor of 5.90×. Our results highlight the importance of analyzing and optimizing task granularity on the Java Virtual Machine.

AutoBoT: Resilient and Cost-Effective Scheduling of a Bag of Tasks on Spot VMs

Many data and task parallel applications can be modeled as a Bag of Tasks (BoT), and scheduled on distributed systems such as Grids, Clusters, and Clouds. We propose AutoBoT, a collection of scheduling strategies for BoTs with hard deadlines on Cloud Virtual Machines (VMs), to lower the overall monetary cost – a distinctive factor for Clouds. Besides reliable fixed-price VMs, AutoBoT uniquely reduces costs by including preemptible spot-priced VMs that are much cheaper, but are unreliable and have time-variant pricing. It guarantees timely completion by making active runtime decisions on pricing, number of VMs to acquire/release, and on task placement, checkpointing and migration. Our rigorous simulations of 7 Million BoT runs sampled from the Google cluster workload uses a realistic Cloud model and $6~months$6months of Amazon EC2 pricing data to compare AutoBoT against two baseline algorithms. We analyze the impact of BoT size, data centers, time periods, deadline duration, loss budget and checkpointing strategies. AutoBoT often gives $\approx$≈80% profit and rare but bounded losses, compared to using only fixed-price VMs. Further, its 100 percent completion guarantee is 23-42 percent better than using only spot-priced VMs which offer a similar profit.

Revisiting conventional task schedulers to exploit asymmetry in multi-core architectures for dense linear algebra operations

Abstract Dealing with asymmetry in the architecture opens a plethora of questions related with the performance- and energy-efficient scheduling of task-parallel applications. While there exist early attempts to tackle this problem, for example via ad-hoc strategies embedded in a runtime framework, in this paper we take a different path, which consists in addressing the asymmetry at the library-level by developing a few asymmetry-aware fundamental kernels. The appealing consequence is that the architecture heterogeneity remains then hidden from the task scheduler. In order to illustrate the advantage of our approach, we employ two well-known matrix factorizations, key to the solution of dense linear systems of equations. From the perspective of the architecture, we consider two low-power processors, one of them equipped with ARM big.LITTLE technology; furthermore, we include in the study a different scenario, in which the asymmetry arises when the cores of an Intel Xeon server operate at two distinct frequencies. For the specific domain of dense linear algebra, we show that dealing with asymmetry at the library-level is not only possible but delivers higher performance than a naive approach based on an asymmetry-oblivious scheduler. Furthermore, this solution is also competitive in terms of performance compared with an ad-hoc asymmetry-aware scheduler furnished with sophisticated scheduling techniques.

NUMA-aware scheduling and memory allocation for data-flow task-parallel applications

Dynamic task parallelism is a popular programming model on shared-memory systems. Compared to data parallel loop-based concurrency, it promises enhanced scalability, load balancing and locality. These promises, however, are undermined by non-uniform memory access (NUMA) systems. We show that it is possible to preserve the uniform hardware abstraction of contemporary task-parallel programming models, for both computing and memory resources, while achieving near-optimal data locality. Our run-time algorithms for NUMA-aware task and data placement are fully automatic, application-independent, performance-portable across NUMA machines, and adapt to dynamic changes. Placement decisions use information about inter-task data dependences and reuse. This information is readily available in the run-time systems of modern task-parallel programming frameworks, and from the operating system regarding the placement of previously allocated memory. Our algorithms take advantage of data-flow style task parallelism, where the privatization of task data enhances scalability through the elimination of false dependences and enables fine-grained dynamic control over the placement of application data. We demonstrate that the benefits of dynamically managing data placement outweigh the privatization cost, even when comparing with target-specific optimizations through static, NUMA-aware data interleaving. Our implementation and the experimental evaluation on a set of high-performance benchmarks executing on a 192-core system with 24 NUMA nodes show that the fraction of local memory accesses can be increased to more than 99%, resulting in a speedup of up to 5× compared to a NUMA-aware hierarchical work-stealing baseline.

Toward Sci-φ: A Lightweight Cloud PaaS for Developing Embarrassingly Parallel Applications Based on Jini

Embarrassingly parallel problems are characterised by a very small amount of information to be exchanged among the parts they are split in, during their parallel execution. As a consequence they do not require sophisticated, low-latency, high-bandwidth interconnection networks but can be efficiently computed in parallel by exploiting commodity hardware. Basically, this means cheap clusters, networks of workstations and desktops, and Computational Clouds. This computational model can be exploited to compute a quite large range of problems. This paper describes Sci-φ, an almost complete redesign of a previous tool of ours aimed at developing task parallel applications based on Java and Jini that were shown to be an effective and efficient solution in environments like clusters and networks of workstations and desktops.

Characterizing and Mitigating Work Time Inflation in Task Parallel Programs

Task parallelism raises the level of abstraction in shared memory parallel programming to simplify the development of complex applications. However, task parallel applications can exhibit poor performance due to thread idleness, scheduling overheads, andwork time inflation– additional time spent by threads in a multithreaded computation beyond the time required to perform the same work in a sequential computation. We identify the contributions of each factor to lost efficiency in various task parallel OpenMP applications and diagnose the causes of work time inflation in those applications. Increased data access latency can cause significant work time inflation in NUMA systems. Our locality framework for task parallel OpenMP programs mitigates this cause of work time inflation. Our extensions to the Qthreads library demonstrate that locality-aware scheduling can improve performance up to 3X compared to the Intel OpenMP task scheduler.

On cluster resource allocation for multiple parallel task graphs

Many scientific applications can be structured as parallel task graphs (PTGs), that is, graphs of data-parallel tasks. Adding data parallelism to a task-parallel application provides opportunities for higher performance and scalability, but poses additional scheduling challenges. In this paper, we study the off-line scheduling of multiple PTGs on a single, homogeneous cluster. The objective is to optimize performance without compromising fairness among the PTGs. We consider the range of previously proposed scheduling algorithms applicable to this problem, from both the applied and the theoretical literature, and we propose minor improvements when possible. Our main contribution is an extensive evaluation of these algorithms in simulation, using both synthetic and real-world application configurations, using two different metrics for performance and one metric for fairness. We identify a handful of algorithms that provide good trade-offs when considering all these metrics. The best algorithm overall is one that structures the schedule as a sequence of phases of increasing duration based on a makespan guarantee produced by an approximation algorithm.

Extreme-scale scripting: Opportunities for large task-parallel applications on petascale computers

Parallel scripting is a loosely-coupled programming model in which applications are composed of highly parallel scripts of program invocations that process and exchange data via files. We characterize here the applications that can benefit from parallel scripting on petascale-class machines, describe the mechanisms that make this feasible on such systems, and present results achieved with parallel scripts on currently available petascale computers.

COHESION — A microkernel based Desktop Grid platform for irregular task-parallel applications

We present Cohesion, a novel approach to Desktop Grid Computing. A major design goal of Cohesion is to enable advanced parallel programming models and application specific frameworks. We focus on methods for irregularly structured task-parallel problems, which require fully dynamic problem decomposition. Cohesion overcomes limitations of classical Desktop Grid platforms by employing peer-to-peer principles and a flexible system architecture based on a microkernel approach. Arbitrary modules can be dynamically loaded to replace default functionality, resulting in a platform that can easily adapt to application specific requirements. We discuss two representative example applications and report on the results of performance experiments that especially consider the high volatility of resources prevailing in a Desktop Grid.

Scheduling Task Parallel Applications for Rapid Turnaround on Enterprise Desktop Grids

Desktop Grids are popular platforms for high throughput applications, but due to their inherent resource volatility it is difficult to exploit them for applications that require rapid turnaround. Efficient desktop Grid execution of short-lived applications is an attractive proposition and we claim that it is achievable via intelligent resource selection. We propose three general techniques for resource selection: resource prioritization, resource exclusion, and task duplication. We use these techniques to instantiate several scheduling heuristics. We evaluate these heuristics through trace-driven simulations of four representative desktop Grid configurations. We find that ranking desktop resources according to their clock rates, without taking into account their availability history, is surprisingly effective in practice. Our main result is that a heuristic that uses the appropriate combination of resource prioritization, resource exclusion, and task replication can achieve performance within a factor of 1.7 of optimal in practice.

Characterizing resource availability in enterprise desktop grids

Desktop grids, which use the idle cycles of many desktop PCs, are one of the largest distributed systems in the world. Despite the popularity and success of many desktop grid projects, the heterogeneity and volatility of hosts within desktop grids have been poorly understood. Yet, resource characterization is essential for accurate simulation and modelling of such platforms. In this paper, we present application-level traces of four real desktop grids that can be used for simulation and modelling purposes. In addition, we describe aggregate and per host statistics that reflect the heterogeneity and volatility of desktop grid resources. Finally, we apply our characterization to develop a performance model for desktop grid applications for various task granularities, and then use a cluster equivalence metric to quantify the utility of the desktop grid relative to that of a dedicated cluster for task-parallel applications.

Implementation of Fault-Tolerant GridRPC Applications

A task parallel application is implemented with Ninf-G, a GridRPC system. A series of experiments are conducted on the Grid testbed in Asia Pacific for three months. Through tens of long executions, typical fault patterns were collected, and instability of the network throughput was determined to be a major reason of the faults. Several important points are stressed to avoid task throughput decline due to the fault-recovery operations: Timeout minimization for fault detection, background recovery, redundant task assignments, and so on. This study also issues a steer for design of the automated fault-tolerant mechanism in an upper layer of the GridRPC framework.

Visualization Aided Performance Tuning of Irregular Task-Parallel Computations

This paper deals with a visualization-based approach to performance analyzing and tuning of highly irregular task-parallel applications. At its core lies a novel automatic layout algorithm for execution graphs which is based on Sugiyama's framework. Our visualization enables the application designer to reliably detect manifestations of parallel overhead and to investigate on their individual root causes. We particularly focus on structural properties of task-parallel computations which are hard to detect in a more analytical way, for example, false sharing and false parallelism. In addition, we discuss embedding our visualization into an integrated development environment, realizing a seamless work-flow for implementation, execution, analysis, and tuning of parallel programs.

Mixed data and task parallelism with HPF and PVM

We present a framework to design efficient and portable HPF applications which exploit a mixture of task and data parallelism. According to the framework proposed, data parallelism is restricted within HPF modules, and task parallelism is achieved by the concurrent execution of several data-parallel modules cooperating through COLTHPF, a coordination layer implemented on top of PVM. COLTHPF can be used independently of the HPF compilation system exploited, and it allows instances of cooperating HPF tasks to be created either statically or at run-time. We claim that COLTHPF can be exploited by means of a simple skeleton-based coordination language and associated compiler to easily express mixed data and task parallel applications runnable on either multicomputers or cluster of workstations. We used a physics application as a test case of our approach for mixing task and data parallelism, and we present the results of several experiments conducted on a cluster of Linux SMPs.

Scheduling parallel applications in distributed networks

Prophet is a run-time scheduling system designed to support the efficient execution of parallel applications written in the Mentat programming language (Grimshaw, 1993). Prior results demonstrated that SPMD applications could be scheduled automatically in an ethernet-based local-area workstation network with good performance (Weissman and Grimshaw, 1994 and 1995). This paper describes our recent efforts to extend Prophet along several dimensions: improved overhead control, greater resource sharing, greater resource heterogeneity, wide-area scheduling, and new application types. We show that both SPMD and task parallel applications can be scheduled effectively in a shared heterogeneous LAN environment containing ethernet and ATM networks by exploiting the application structure and dynamic run-time information.