This paper presents an in-depth analysis of the impact of system noise on large-scale parallel application performance in realistic settings. Our analytical model shows that not only collective operations but also point-to-point communications influence the application's sensitivity to noise. We present a simulation toolchain that injects noise delays from traces gathered on common large-scale architectures into a LogGPS simulation and allows new insights into the scaling of applications in noisy environments. We investigate collective operations with up to 1 million processes and three applications (Sweep3D, AMG, and POP) with up to 32,000 processes.We show that the scale at which noise becomes a bottleneck is system-specific and depends on the structure of the noise. Simulations with different network speeds show that a 10x faster network does not improve application scalability. We quantify noise and conclude that our tools can be utilized to tune the noise signatures of a specific system.

/pdf/characterizing-the-influence-of-system-noise-on-large-scale-4li7sabdt2.pdf

Characterizing the Influence of System Noise on Large-Scale Applications by Simulation

Operating system noise has been shown to be a key limiter of application scalability in high-end systems. While several studies have attempted to quantify the sources and effects of system interference using user-level mechanisms, there are few published studies on the effect of different kinds of kernel-generated noise on application performance at scale. In this paper, we examine the sensitivity of real-world, large-scale applications to a range of OS noise patterns using a kernel-based noise injection mechanism implemented in the Catamount lightweight kernel. Our results demonstrate the importance of how noise is generated, in terms of frequency and duration, and how this impact changes with application scale. For example, our results show that 2.5% net processor noise at 10,000 nodes can have no impact or can result in over a factor of 20 slowdown for the same application, depending solely on how the noise is generated. We also discuss how the characteristics of the applications we studied, for example computation/communication ratios, collective communication sizes, and other characteristics, related to their tendency to amplify or absorb noise. Finally, we discuss the implications of our findings on the design of new operating systems, middleware, and other system services for high-end parallel systems.

Characterizing application sensitivity to OS interference using kernel-level noise injection

High performance cloud computing

As parallel jobs get bigger in size and finer in granularity, "system noise" is increasingly becoming a problem. In fact, fine-grained jobs on clusters with thousands of SMP nodes run faster if a processor is intentionally left idle (per node), thus enabling a separation of "system noise" from the computation. Paying a cost in average processing speed at a node for the sake of eliminating occasional processes delays is (unfortunately) beneficial, as such delays are enormously magnified when one late process holds up thousands of peers with which it synchronizes.We provide a probabilistic argument showing that, under certain conditions, the effect of such noise is linearly proportional to the size of the cluster (as is often empirically observed). We then identify a major source of noise to be indirect overhead of periodic OS clock interrupts ("ticks"), that are used by all general-purpose OSs as a means of maintaining control. This is shown for various grain sizes, platforms, tick frequencies, and OSs. To eliminate such noise, we suggest replacing ticks with an alternative mechanism we call "smart timers". This turns out to also be in line with needs of desktop and mobile computing, increasing the chances of the suggested change to be accepted.

/pdf/system-noise-os-clock-ticks-and-fine-grained-parallel-5bexb81vi3.pdf

System noise, OS clock ticks, and fine-grained parallel applications

Predictable performance is important for understanding and alleviating application performance issues; quantifying the effects of source code, compiler, or system software changes; estimating the time required for batch jobs; and determining the allocation requests for proposals. Our experiments show that on a Cray XE system, the execution time of a communication-heavy parallel application ranges from 28% faster to 41% slower than the average observed performance. Blue Gene systems, on the other hand, demonstrate no noticeable run-to-run variability. In this paper, we focus on Cray machines and investigate potential causes for performance variability such as OS jitter, shape of the allocated partition, and interference from other jobs sharing the same network links. Reducing such variability could improve overall throughput at a computer center and save energy costs.

/pdf/there-goes-the-neighborhood-performance-degradation-due-to-2bd516y4mz.pdf

There goes the neighborhood: performance degradation due to nearby jobs

A parallel application benefits from scheduling policies that include a global perspective of the application's process working set. As the interactions among cooperating processes increase, mechanisms to ameliorate waiting within one or more of the processes become more important. In particular, collective operations such as barriers and reductions are extremely sensitive to even usually harmless events such as context switches among members of the process working set. For the last 18 months, we have been researching the impact of random short-lived interruptions such as timer-decrement processing and periodic daemon activity, and developing strategies to minimize their impact on large processor-count SPMD bulk-synchronous programming styles. We present a novel co-scheduling scheme for improving performance of fine-grain collective activities such as barriers and reductions, describe an implementation consisting of operating system kernel modifications and run-time system, and present a set of empirical results comparing the technique with traditional operating system scheduling. Our results indicate a speedup of over 300% on synchronizing collectives.

Improving the Scalability of Parallel Jobs by adding Parallel Awareness to the Operating System

In a parallel computing environment comprising a network of SMP nodes each having at least one processor, a parallel-aware co-scheduling method and system for improving the performance and scalability of a dedicated parallel job having synchronizing collective operations. The method and system uses a global co-scheduler and an operating system kernel dispatcher adapted to coordinate interfering system and daemon activities on a node and across nodes to promote intra-node and inter-node overlap of said interfering system and daemon activities as well as intra-node and inter-node overlap of said synchronizing collective operations. In this manner, the impact of random short-lived interruptions, such as timer-decrement processing and periodic daemon activity, on synchronizing collective operations is minimized on large processor-count SPMD bulk-synchronous programming styles.

Jeffrey Fier

Papers

Improving the Scalability of Parallel Jobs by adding Parallel Awareness to the Operating System

Parallel-aware, dedicated job co-scheduling method and system

Parallel-aware, dedicated job co-scheduling within/across symmetric multiprocessing nodes