Automatic I/O Scheduler Selection through Online Workload Analysis

Future Generation Computer Systems

et al. 2020

Self Cite

I/O optimization techniques such as request scheduling can improve performance mainly for the access patterns they target, or they depend on the precise tune of parameters. In this paper, we propose an approach to adapt the I/O forwarding layer of HPC systems to the application access patterns by tuning a request scheduler. Our case study is the TWINS scheduling algorithm, where performance improvements depend on the time window parameter, which depends on the current workload. Our approach uses a reinforcement learning technique-contextual bandits-to make the system capable of learning the best parameter value to each access pattern during its execution, without a previous training phase. We evaluate our proposal and demonstrate it can achieve a precision of 88% on the parameter selection in the first hundreds of observations of an access pattern. After having observed an access pattern for a few minutes (not necessarily contiguously), we demonstrate that the system will be able to optimize its performance for the rest of the life of the system (years).

Section: Related Workmentioning

confidence: 99%

Adaptive request scheduling for the I/O forwarding layer using reinforcement learning

Bez

Boito

Future Generation Computer Systems

et al. 2020

Self Cite

“…A similar pattern matching strategy was used with success in a previous work [22] to detect patterns of disk block accesses, where each access is represented by its logical block number. However, in our scenario requests are not to blocks but at file level -they request a portion of data of a given size, from a given file, starting at a given offset.…”

Section: File Access Pattern Matchingmentioning

confidence: 99%

On server-side file access pattern matching

Boito

2019 International Conference on High Performance Computing &Amp; Simulation (HPCS)

Pilla

et al. 2019

Self Cite

In this paper, we propose a pattern matching approach for server-side access pattern detection for the HPC I/O stack. More specifically, our proposal concerns file-level accesses, such as the ones made to I/O libraries, I/O nodes, and the parallel file system servers. The goal of this detection is to allow the system to adapt to the current workload. Compared to existing detection techniques, ours differ by working at run-time and on the server side, where detailed application information is not available since HPC I/O systems are stateless, and without relying on previous traces. We build a time series to represent accesses spatiality, and use a pattern matching algorithm, in addition to an heuristic, to compare it to known patterns. We detail our proposal and evaluate it with two case studies-situations where detecting the current access pattern is important to select the best scheduling algorithm or to tune a fixed algorithm parameter. We show our approach has good detection capabilities, with precision of up to 93% and recall of up to 99%, and discuss all design choices.

“…The onloaded service selected for the evaluation is an I/O workflow analyzer to select the I/O scheduler (IOAnalyzer [12]) for the current workload automatically. As the service is highly optimized, we modified it in order to increase its CPU consumption.…”

Section: Dyonmentioning

confidence: 99%

“…This extra capacity can be used, for instance, to reduce a kernel critical path by executing code in parallel, or to run additional services that provide additional benefits. Consider an I/O analyser as IOAnalyzer [12] (to improve the I/O scheduler) as an example of service (we will henceforth call them Onloaded Services in general). This service can reduce the I/O time of a job with some I/O phases up to 5 times [12] if the disk scheduler becomes optimal.…”

Section: Introductionmentioning

confidence: 99%

“…Consider an I/O analyser as IOAnalyzer [12] (to improve the I/O scheduler) as an example of service (we will henceforth call them Onloaded Services in general). This service can reduce the I/O time of a job with some I/O phases up to 5 times [12] if the disk scheduler becomes optimal. This benefit affects the rest of the system, which can increase its performance or go to an IDLE state faster.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

DYON: Managing a New Scheduling Class to Improve System Performance in Multicore Systems

Euro-Par 2013: Parallel Processing Workshops

Giralt

Cortes

2014

Self Cite

Abstract. Due to the increase in the number of available cores in current systems, plenty of system software starts to use some of these cores to perform tasks that will help optimize the application behaviour. Unfortunately, current Onload mechanisms are too limited. On the one hand, there is no dynamic way to decide the number of cores that is taken from applications and given to these system helpers. And, on the other hand, the onload mechanisms do not offer enough control over when and where onloading tasks should to be executed. In this paper we propose a new Onload Framework that addresses these issues. First, we propose DYON, a dynamic and adaptive method to control the amount of extra CPUs offered to the Onload Framework to generate benefits for the whole system. And second, we propose a submission mechanism that given a task, executes it if there are idle resources or rejects it otherwise. This feature is useful to move the execution of small pieces of code out of the critical path (allowing parallel execution) when this is possible, or discard them and execute a code that will not rely on them.