Heuristics for scheduling I/O operations

The I/O bottleneck in parallel computer systems has recently begun receiving increasing interest. Most attention has focused on improving the performance of I/O devices using fairly low level parallelism in techniques such as disk striping and interleaving. Widely applicable solutions, however, will require an integrated approach which addresses the problem at multiple system levels, including applications, systems software, and architecture. We propose that within the context of such an integrated approach, scheduling parallel I/O operations will become increasingly attractive and can potentially provide substantial performance benefits. We describe a simple I/O scheduling problem and present approximate algorithms for its solution. The costs of using these algorithms in terms of execution time, and the benefits in terms of reduced time to complete a batch of I/O operations, are compared with the situations in which no scheduling is used, and in which an optimal scheduling algorithm is used. The comparison is performed both theoretically and experimentally. We have found that, in exchange for a small execution time overhead, the approximate scheduling algorithms can provide substantial improvements in I/O completion times     

高性能计算环境下基于远程I/O负载平衡调度算法

在高性能计算环境中数据和应用程序往往分散在不同的节点,远程I/O的效率成为影响高性能计算性能的一个重要因素。为了提高系统的I/O效率,引入了一种基于远程I/O负载平衡调度算法。该算法采用了一种预约机制,可以对节点I/O负载进行动态调整、更好地利用网络带宽。详细介绍了该算法的实现过程,并且在一个模拟环境下对该算法的效果进行了评测。     

MapReduce连接查询的I/O代价研究

数据的指数级增长给数据管理和分析带来了严峻的挑战.连接查询是数据分析中一种常用运算,而MapReduce是一种用于大规模数据集并行处理的编程模型,研究基于MapReduce的连接查询代价评估和查询优化,有着学术意义和应用价值.MapReduce连接查询算法的性能主要取决于I/O代价(包括本地和网络I/O),而I/O代价与数据集以及连接运算的特征参数相关,通过对二元连接的I/O代价评估可以优化多元连接执行计划.基于此,首先提出了二元连接查询的I/O代价模型;随后,对现有二元连接算法进行形式化定义和简单扩展,归纳出6种基于MapReduce连接查询算法,并通过算法白盒分析定义它们的I/O代价函数;最后,提出一种多元连接最优执行计划的选择算法.通过实验表明I/O代价模型的正确性且能够准确地反映算法的性能优劣.     

Data movement optimizations for independent MPI I/O on the Blue Gene/Q

Scalable high-performance I/O is crucial for application performance on large-scale systems. With the growing complexity of the system interconnects, it has become important to consider the impact of network contention on I/O performance because the I/O messages traverse several hops in the interconnect before reaching the I/O nodes or the file system. In this work, we present a route-aware and load-aware algorithm to modify existing bridge node assignment in the Blue Gene/Q (BG/Q) supercomputer. We reduce the network contention and reduce the write time by an average of 60% over the default independent I/O and by 20% over collective I/O on up to 8192 nodes on the Mira BG/Q system. Our algorithm routes 1.4× fewer messages through the bridge nodes which connect to the I/O nodes on the BG/Q. Our algorithm also reduces the average distance of a compute node from a bridge node, and thus lessens the network load, and decreases I/O time.     

Tight bounds for prefetching and buffer management algorithms forparallel I/O systems

The I/O performance of applications in multiple-disk systems can be improved by overlapping disk accesses. This requires the use of appropriate prefetching and buffer management algorithms that ensure the most useful blocks are accessed and retained in the buffer. In this paper, we answer several fundamental questions on prefetching and buffer management for distributed-buffer parallel I/O systems. First, we derive and prove the optimality of an algorithm, P-min, that minimizes the number of parallel I/Os. Second, we analyze P-con, an algorithm that always matches its replacement decisions with those of the well-known demand-paged MIN algorithm. We show that P-con can become fully sequential in the worst case. Third, we investigate the behavior of on-line algorithms for multiple-disk prefetching and buffer management. We define and analyze P-Iru, a parallel version of the traditional LRU buffer management algorithm. Unexpectedly, we find that the competitive ratio of P-Iru is independent of the number of disks. Finally, we present the practical performance of these algorithms on randomly generated reference strings. These results confirm the conclusions derived from the analysis on worst case inputs     

Formal derivation of concurrent programs: an example from industry

The formal derivation of an implementation of the I/O (input/output) subsystem portion of an existing operating system is presented. The I/O subsystem is responsible for allocating I/O resources such as tapes, disks, I/O channels in response to requests from user processes. The derivation employs the UNITY methodology which captures the concurrent interaction of the I/O subsystem with its environment. The verified resource allocation algorithm that results from the derivation has been used as part of a high-level design by software engineers implementing the I/O subsystem. As the largest application to date of the UNITY methodology, the derivation illustrates a number of techniques for organizing large specifications and proofs     

Energy-aware disk scheduling for soft real-time I/O requests

In this work, we develop energy-aware disk scheduling algorithm for soft real-time I/O. Energy consumption is one of the major factors which bar the adoption of hard disk in mobile environment. Heat dissipation of large scale storage system also calls for an energy-aware scheduling technique to further increase the storage density. The basic idea in this work is to properly determine the I/O burst size so that device can be in standby mode between consecutive I/O bursts and that it can satisfy the soft real-time requirement. We develop an elaborate model which incorporates the energy consumption characteristics, overhead of mode transition in determining the appropriate I/O burst size and the respective disk operating schedule. Efficacy of energy-aware disk scheduling algorithm greatly relies on not only disk scheduling algorithm itself but also various operating system and device firmware related concerns. It is crucial that the various operating system level and device level features need to be properly addressed within disk scheduling framework. Our energy-aware disk scheduling algorithm successfully addresses a number of outstanding issues. First, we examine the effect of OS and hard disk firmware level prefetch policy and incorporate its effect in our disk scheduling framework. Second, our energy aware scheduling framework can allocate a certain fraction of disk bandwidth to handle sporadically arriving non real-time I/O’s. Third, we examine the relationship between lock granularity of the buffer management and energy consumption. We develop a prototype software with energy-aware scheduling algorithm. In our experiment, proposed algorithm can reduce the energy consumption to one fourth if we use energy-aware disk scheduling algorithm. However, energy-aware disk scheduling algorithm increases buffer requirement significantly, e.g., from 4 to 140 KByte. We carefully argue that the buffer overhead is still justifiable given the cost of DRAM chip and importance of energy management in modern mobile devices. The result of our work not only provides the energy efficient scheduling algorithm but also provides an important guideline in capacity planning of future energy efficient mobile devices. This paper is funded by KOSEF through Statistical Research Paper for Complex System at Seoul National University.     

一种嵌入式自适应调度算法的设计

根据嵌入式实时系统的需求及应用领域中任务的特点,提出一种针对两种常见的不同类型任务,进行自适应调度算法ERTAS（Real-Time Embedded Adaptive Scheduling Algorithm,并提出了其系统结构。分析μC/OS-Ⅱ内核中与调度相关的数据结构,提出在其内核中自适应调度算法的实现,主要是包括控制器与基础调度器的实现。通过仿真实验比较了改进的设计算法与原调度算法的性能,通过分析说明了该调度算法的可行性。     

STXXL: standard template library for XXL data sets

We present the software library STXXL that is an implementation of the C++ standard template library (STL) for processing huge data sets that can fit only on hard disks. It supports parallel disks, overlapping between disk I/O and computation and it is the first I/O‐efficient algorithm library that supports the pipelining technique that can save more than half of the I/Os. STXXL has been applied both in academic and industrial environments for a range of problems including text processing, graph algorithms, computational geometry, Gaussian elimination, visualization, and analysis of microscopic images, differential cryptographic analysis, etc. The performance of STXXL and its applications are evaluated on synthetic and real‐world inputs. We present the design of the library, how its performance features are supported, and demonstrate how the library integrates with STL. Copyright © 2007 John Wiley & Sons, Ltd.     

磁盘缓存管理机制研究

磁盘缓存是解决I/O性能的一种技术。文章主要讲述缓存管理组成、算法的种类及其管理策略。并对基于频率的替换算法的原理、实现方法做了详细阐述。     

基于闪存的混合存储系统缓冲区管理算法

由于基于闪存的混合存储系统充分利用了闪存的高速随机读和磁盘的快速顺序写的特性,近年来已经成为了数据库管理系统的二级存储层的高效存储模式,但其I/O访问开销是一个继续提高存储性能的瓶颈.为了降低混合存储系统的I/O访问开销,提出了一种自适应缓冲区管理算法DLSB.该算法根据数据页的逻辑代价和物理代价进行自适应的数据域选择;并在选择的数据域中,比较闪存队列和磁盘队列容量的实际值与理想值来确定数据页的置换,达到了提高I/O访问效率的目的.实验结果表明,该算法有效且可行,显著降低了混合存储系统的I/O访问开销.     

基于Reed-Solomon算法的RAID机制的设计与实现

为了满足海量信息存储可靠性的要求，提出了把Reed Solomon算法应用到RAID系统中的方法，并给出了在Linux环境下系统实现的方案RSRAID。通过对系统性能及可靠性进行测试，并与其他RAID机制进行对比，证明系统具有良好的I/O性能和更高的可靠性。     

Two-dimensional substring indexing

As databases have expanded in scope to storing string data (XML documents, product catalogs), it has become increasingly important to search databases based on matching substrings, often on multiple, correlated dimensions. While string B-trees are I/O optimal in one dimension, no index structure with non-trivial query bounds is known for two-dimensional substring indexing. In this paper, we present a technique for two-dimensional substring indexing based on a reduction to the geometric problem of identifying common colors in two ranges containing colored points. We develop an I/O efficient algorithm for solving the common colors problem, and use it to obtain an I/O efficient (poly-logarithmic query time) algorithm for the two-dimensional substring indexing problem. Our techniques result in a family of secondary memory index structures that trade space for time, with no loss of accuracy.     

Techniques for Scheduling I/O in a High Performance Multimedia-on-Demand Server

One of the key components of a multiuser multimedia-on-demand system is the data server. Digitalization of traditionally analog data such as video and audio, and the feasibility of obtaining network bandwidths above the gigabit-per-second range, are two important advances that have made possible the realization, in the near future, of interactive distributed multimedia systems. Secondary-to-main memory I/O technology has not kept pace with advances in networking, main memory, and CPU processing power. Consequently, the performance of the server has a direct bearing on the overall performance of such a system. In this paper, we present a highperformance solution to the I/O retrieval problem in a distributed multimedia system. We develop a model for the architecture of a server for such a system. Parallelism of data retrieval is achieved by striping the data across multiple disks. We present the algorithms for server operation when servicing a constant number of streams, as well as the admission control policy for accepting requests for new streams. The performance of any server ultimately depends on the data access patterns. Two modifications of the basic retrieval algorithm are presented to exploit data access patterns in order to improve system throughput and response time. Finally, we present preliminary performance results of these algorithms on the IBM SP1 and Intel Paragon parallel computers.     

一种动态优先级排序的虚拟机I/O调度算法

I/O任务调度是影响I/O密集型虚拟机性能的重要因素。现有调度方法主要是针对虚拟机整机I/O带宽的优化,较少兼顾各虚拟域与全局性能,也无法满足域间差异化服务的要求。针对现有方法的不足,提出了一种动态优先级排序的虚拟机I/O调度算法DPS。该算法基于多属性决策理论,以离差最大化方法计算I/O任务的优先级评估属性权重,对I/O任务优先级进行综合评估;通过引入任务所在虚拟域价值,体现云计算环境下虚拟域重要性差异。在Xen系统中通过实验评测DPS调度虚拟化网卡的性能,结果表明,DPS能够有效提高指定域与全局的I/O任务截止期保证率、整机I/O带宽,并能为不同虚拟域的I/O应用提供差异化服务。     

Slabpose Columnsort: A New Oblivious Algorithm for Out-of-Core Sorting on Distributed-Memory Clusters

Our goal is to develop a robust out-of-core sorting program for a distributed-memory cluster. The literature contains two dominant paradigms for out-of-core sorting algorithms: merging-based and partitioning-based. We explore a third paradigm, that of oblivious algorithms. Unlike the two dominant paradigms, oblivious algorithms do not depend on the input keys and therefore lead to predetermined I/O and communication patterns in an out-of-core setting. Predetermined I/O and communication patterns facilitate overlapping I/O, communication, and computation for efficient implementation. We have developed several out-of-core sorting programs using the paradigm of oblivious algorithms. Our baseline implementation, 3-pass columnsort, was based on Leighton's columnsort algorithm. Though efficient in terms of I/O and communication, 3-pass columnsort has a restriction on the maximum problem size. As our first effort toward relaxing this restriction, we developed two implementations: subblock columnsort and M-columnsort. Both of these implementations incur substantial performance costs: subblock columnsort performs additional disk I/O, and M-columnsort needs substantial amounts of extra communication and computation. In this paper we present slabpose columnsort, a new oblivious algorithm that we have designed explicitly for the out-of-core setting. Slabpose columnsort relaxes the problem-size restriction at no extra I/O or communication cost. Experimental evidence on a Beowulf cluster shows that unlike subblock columnsort and M-columnsort, slabpose columnsort runs almost as fast as 3-pass columnsort. To the best of our knowledge, our implementations are the first out-of-core multiprocessor sorting algorithms that make no assumptions about the keys and produce output that is perfectly load balanced and in the striped order assumed by the Parallel Disk Model.     

Xen的I/O虚拟化性能分析与优化

在Credit算法应用中,由I/O事务唤醒的VCPU处于最高优先级BOOST状态,优先抢占PCPU资源,使I/O操作的响应速度提高,但多个虚拟机同时进行I/O操作时,会引起较长延时和公平性原则被破坏问题.针对这个问题,研究分析SEDF算法、Credit算法、Credit2算法,提出L-Credit调度算法解决多个虚拟机同时进行I/O操作引起响应延迟的问题.通过监测I/O设备环共享页面中响应和请求的个数的方法,对处于BOOST状态下I/O操作进一步细化排序,使稀疏型I/O操作较密集型I/O操作先调用执行.通过对L-Credit算法与Credit算法在同一应用场景下反复对比实验,得出L-Credit算法可以提高I/O响应性能,并且继承了Credit算法负载均衡和按比例公平共享的特点.     

Efficient disk-based K-means clustering for relational databases

K-means is one of the most popular clustering algorithms. We introduce an efficient disk-based implementation of K-means. The proposed algorithm is designed to work inside a relational database management system. It can cluster large data sets having very high dimensionality. In general, it only requires three scans over the data set. It is optimized to perform heavy disk I/O and its memory requirements are low. Its parameters are easy to set. An extensive experimental section evaluates quality of results and performance. The proposed algorithm is compared against the Standard K-means algorithm as well as the Scalable K-means algorithm.     

Block I/O Scheduling on Storage Servers of Distributed File Systems

This paper presents a new scheme of I/O scheduling on storage servers of distributed/parallel file systems, for yielding better I/O performance. To this end, we first analyze read/write requests in the I/O queue of storage server (we name them block I/Os), by using our proposed technique of horizontal partition. Then, all block requests are supposed to be divided into multiple groups, on the basis of their offsets. This is to say, all requests related to the same chunk file will be grouped together, and then be satisfied within the same time slot between opening and closing the target chunk file on the storage server. As a result, the time resulted by completing block I/O requests can be significantly decreased, because of less file operations on the corresponding chunk files at the low-level file systems of server machines. Furthermore, we introduce an algorithm to rate a priority for each group of block I/O requests, and then the storage server dispatches groups of I/Os by following the priority order. Consequently, the applications having higher I/O priorities, e.g. they have less I/O operations and small size of involved data, can finish at a earlier time. We implement a prototype of this server-side scheduling in the PARTE file system, to demonstrate the feasibility and applicability of the proposed scheme. Experimental results show that the newly proposed scheme can achieve better I/O bandwidth and less I/O time, compared with the strategy of First Come First Served, as well as other server-side I/O scheduling approaches.     

Efficient Algorithms for Mining and Incremental Update of Maximal Frequent Sequences

We study two problems: (1) mining frequent sequences from a transactional database, and (2) incremental update of frequent sequences when the underlying database changes over time. We review existing sequence mining algorithms including GSP, PrefixSpan, SPADE, and ISM. We point out the large memory requirement of Pref ixSpan, SPADE, and ISM, and evaluate the performance of GSP. We discuss the high I/O cost of GSP, particularly when the database contains long frequent sequences. To reduce the I/O requirement, we propose an algorithm MFS, which could be considered as a generalization of GSP. The general strategy of MFS is to first find an approximate solution to the set of frequent sequences and then perform successive refinement until the exact set of frequent sequences is obtained. We show that this successive refinement approach results in a significant improvement in I/O cost. We discuss how MFS can be applied to the incremental update problem. In particular, the result of a previous mining exercise can be used (by MFS) as a good initial approximate solution for the mining of an updated database. This results in an I/O efficient algorithm. To improve processing efficiency, we devise pruning techniques that, when coupled with GSP or MFS, result in algorithms that are both CPU and I/O efficient.This research is supported by Hong Kong Research Grants Council grant HKU 7040/02E.