I/O Scheduling Service for Multi-Application Clusters

Abstract

Distributed applications, especially the ones being I/O intensive, often access the storage subsystem in a non-sequential way (stride requests). Since such behaviors lower the overall system performance, many applications use parallel I/O libraries such as ROMIO to gather and reorder requests. In the meantime, as cluster usage grows, several applications are often executed concurrently, competing for access to storage subsystems and, thus, potentially canceling optimizations brought by Parallel I/O libraries.The aIOLi project aims at optimizing the I/O accesses within the cluster and providing a simple POSIX API. This article presents an extension of aIOLi to address the issue of disjoint * This work has been done within the ID laboratory jointly supported by CNRS, INPG, INRIA, and UJF and the project LIPS between INRIA and BULL Lab. Computer resources are provided by the grid5000 french experimental grid (http://www.grid5000.fr/). 1 accesses generated by different concurrent applications in a cluster. In such a context, good trade-off has to be assessed between performance, fairness and response time. To achieve this, an I/O scheduling algorithm together with a «requests aggregator» that take into account both application access patterns and global system load, have been designed and merged into aIOLi. This improvement led to the implementation of a new generic framework pluggable into any I/O file system layer. A test composed of two concurrent IOR benchmarks showed improvements on read accesses by a factor ranging from 3.5 to 35 with POSIX calls and from 3.3 to 5 with ROMIO, both reference benchmarks executed on a traditional NFS server without any additional optimizations.

Omnisc'IO: A Grammar-Based Approach to Spatial and Temporal I/O Patterns Prediction

“…This knowledge can indeed help decide which application should be given priority to access the file system given its future access pattern. The scheduler proposed by Lebre et al [35] aims at aggregating and reordering requests while trying to maintain fairness across applications, a task that would undoubtedly be easier with any kind of prediction of future incoming I/O requests.…”

CALCioM: Mitigating I/O Interference in HPC Systems through Cross-Application Coordination

“…Other approaches such as the one from Lebre et al [37] provide multi applications scheduling with the goal of better aggregating and reordering requests, while trying to maintain fairness across applications. The proposed solution does not take into account each application's available resources and required I/O efficiency and does not check the availability of the file system to potentially change the application's behavior.…”

AGIOS: Application-Guided I/O Scheduling for Parallel File Systems

“…As they work on the context of one application, in multi-application scenarios their effects can be impaired by interference caused by concurrent accesses to the file system [5]. I/O scheduling techniques are applied in this situation to coordinate the execution of the requests.…”

“…The algorithm proposed with aIOLi [5] was used with the Lustre file system by Qian et al [13]. They describe the incorporation of the scheduling algorithm in the data servers' code with no global coordination, resulting in performance improvements of up to 40%.…”

“…Our library can be easily integrated into a parallel file system's server code to manage incoming I/O requests using a variation of the Multi-Level Feedback algorithm [5]. This algorithm uses a quantum-based approach to schedule requests coming from different applications, improving performance while trying to keep fairness in the resource sharing and applications' response time.…”