Clock synchronization in high‐end computing environments: a strategy for minimizing clock variance at runtime

Abstract

SUMMARYWe present a new software‐based clock synchronization scheme that provides high precision time agreement among distributed memory nodes. The technique is designed to minimize variance from a reference chimer during runtime and with minimal time‐request latency. Our scheme permits initial unbounded variations in time and corrects both slow and fast chimers (clock skew). An implementation developed within the context of the message passing interface is described, and time coordination measurements are presented. Among our results, the mean time variance for a set of nodes improved from 20.0 ms under standard Network Time Protocol down to 2.29 μs under our scheme. Copyright © 2012 John Wiley & Sons, Ltd.

An efficient virtual system clock for the wireless raspberry pi computer platform

“…Neville observes that depending on the computing platform, the time measurements of the system clock attain errors of several milliseconds even if the resynchronization of the system clock occurs at a short time interval of 16 s between NTP queries to improve the accuracy of system clock. Furthermore, the work on NTP synchronization presented in Jones and Koenig corroborates the experimental results shown in the table.…”

Evaluation of NTP/PTP fine-grain synchronization performance in HPC clusters

“…At the basis of a better usage of the resources, techniques for exploiting more efficient concurrency on applications that run on many cores per chips (i.e., CPUs / Accelerators) and multiple nodes per clusters are of crucial importance [13,14]. Already several works in literature show that an efficient concurrency, which ultimately leads to improvements in TtS and EtS, can be achieved by (i) co-scheduling techniques [13], (ii) optimized coding, which requires application's performance analysis tools for debugging [6,13], and (iii) runtime autotuning techniques, based on power and performance monitoring, to promptly react to workloads changing and events (e.g., by mean of Power Capping and Dynamic Voltage and Frequency Scaling -DVFS; Approximate Computing; advanced cooling control policies) [9,20].…”

“…To address this problem, Jones et al [14] worked on a NTPbased synchronization scheme for MPI applications that provides an accuracy of 2.29 µs (min −32.0 µs, max 32.1 µs, span 64.1 µs). This solution is suitable for runtime synchronization of MPI applications, but not to timestamp and correlate measurements coming from monitoring systems.…”

“…This is especially true for High Performance Computing (HPC) clusters, considering the trend toward increasing the node count to exploit application concurrency and parallelism. In this context, it becomes crucial to ensure a tight level of time agreement [5,13,14]. As reported in [13], this is needed for several reasons, with the ultimate goal to improve the application execution time and energy efficiency of the system (i.e., Time-to-Solution -TtS -and Energy-to-Solution -EtS).…”

“…In particular, within the context of HPC systems and data-centers, NTP is today the most used protocol [13], but -as reported in a recent survey on leadership class supercomputing centers [13] -is usually set with default configurations, obtaining a synchronization accuracy that does not meet acceptable uncertainty requirements for many potential applications (jitter of tens / hundreds of milliseconds). To achieve microseconds synchronization without incurring in extra cost for dedicated hardware support, such as with PTP, several works in literature proposed alternative software solutions [4][5][6]14]. However, the advantages to use robust standards that are widely supported by the community -like NTP and PTPare evident.…”