基于负载特性和服务时间评估改进的AS调度算法

doi:10.11959/j.issn.1000-436x.2015018

摘要/Abstract

摘要：

Linux I/O调度层的预期调度算法AS(anticipatory scheduling)对所有的负载分配相同的预期周期，如果等待的I/O请求没有及时到达，将会带来额外的时延。针对AS算法的不足，提出了一种基于负载特性和服务时间评估改进的AS算法(WPCAS,workload characteristic and service time evluation AS)。WPCAS分为进程归类模块(PC)和服务时间评估模块(STE)两部分。PC模块根据负载特性为每类进程指定不同的预期周期。STE模块根据服务时间决定是否预期下一个请求。通过对比实验表明，WPCAS 在吞吐量、预期成功平均等待时延和伪空闲周期方面都优越于95%-Heuristic和AS算法。

关键词: 存储系统, I/O调度层, AS, WPCAS, 服务时间

Abstract:

Anticipatory scheduling (AS) of Linux I/O schedule layer assigns fixed anticipation period length to each process.It will introduce extra delay if I/O request does not arrive in time.The improved anticipatory scheduling (WPCAS) algorithm is proposed which is based on workload characteristic and service time.WPCAS includes process classifier(PC) module and request service time evaluation (STE) module.PC module assigns different anticipation period lengths in terms of workload characteristic.STE module decides whether to anticipate the coming request according to service time.Experimental results show that WPCAS algorithm is superior to 95%-Heuristic and AS in terms of throughput,average waiting time of anticipation success and deceptive idleness.

Key words: storage system, I/O schedule layer, AS, WPCAS, service time

陈金志,姚念民,蔡绍滨,孙美玲. 基于负载特性和服务时间评估改进的AS调度算法[J]. 通信学报, 2015, 36(1): 159-166.

Jin-zhong CHEN,Nian-min YAO,Shao-bin CAI,Mei-ling SUN. Improved anticipatory scheduling algorithm based on workload characteristic and service time evaluation[J]. Journal on Communications, 2015, 36(1): 159-166.

图/表 9

图1

图2

图3

图4

图5

图6

图7

图8

表1

参考文献 21

[1]	WORTHINGTON B , GANGER G , PATT Y . Scheduling algorithms for modern disk drives[A]. In ACM Sigmetrics[C]. 1994. 241-251.
[2]	HUANG L , CHIUEH T . Implementation of a Rotation Latency Sensitive Disk Scheduler[R]. Technical Report ECSL-TR81,SUNY,Stony Brook. 2000.
[3]	REDDY A L N , WYLLIE J . Disk scheduling in a multimedia I/O system[A]. Proc First ACM Int'l Conf Multimedia (MULTIMEDIA'93)[C]. 1993. 225-233.
[4]	CHANG R I , SHIH W K , CHANG R C . Deadline-modification-scan with maximum scannable-groups for multimedia real-time disk scheduling[A]. Proceedings of the 19th IEEE Real-Time Systems Symposium[C]. 1998. 40-49.
[5]	CHANG H P , CHANG R I , SHIH W K , et al. Reschedulable-group-SCAN scheme for mixed real-time/non-real-time disk scheduling in a multimedia system[J]. J Syst Software, 2002,59(2): 143-152.
[6]	CHANG H P , CHANG R I , SHIH W K , et al. GSR:a global seek-optimizing real-time disk-scheduling algorithm,the journal of transactions in firm real-time database systems[J]. Systems and Software, 2007,80(2): 198-215.
[7]	KESAVAN M , GAVRILOVSKA A , SCHWAN K . Differential virtual time (DVT):rethinking I/O service differentiation for virtual machines[A]. Proc of the 1st ACM Symposium on Cloud Computing,SoCC'10[C].ACM, New York(2010), 2010. 27-38.
[8]	RIZZO L , VALENTE P . Hybrid:achieving deterministic fairness and high throughput in disk scheduling[A]. Proc CCCT '04[C]. New York(2010), 2004.
[9]	WACHS M , ABD-EL-MALEK M , THERESKA E , et al. Argon:performance insulation for shared storage servers[A]. Proc of the 5th USENIX Conference on File and Storage Technologies (FAST'07)[C]. San Jose,CA,USA, 2007. 61-76.
[10]	GULATI A , MERCHANT A , VARMAN P J . Pclock:an arrival curve based approach for QoS guarantees in shared storage systems[J]. SIGMETRICS Performance Evaluation Rev, 2007,35(1): 13-14.
[11]	GULATI A , MERCHANT A , UYSAL M , et al. Efficient and Adaptive Proportional Share I/O Scheduling[R]. Hewlett-Packard Pdf, 2009.
[12]	GANG P , CKER CHIUEH T . Availability and fairness support for storage qos guarantee in distributed computing systems[A]. ICDCS'08.The 28th International Conference on 2008[C]. 2008. 589-596.
[13]	VALENTE P , CHECCONI F . High throughput disk scheduling with fair bandwidth distribution[J]. IEEE Transactions on Computers, 2010,59(9): 1172-1186.
[14]	[EB/OL]. 2010.
[15]	LOVE R . Linux Kernel Development[M]. Developers Library Sams Publishing/Novel, 2005.
[16]	STOUPA K , VAKALI A . QoS-oriented negotiation in disk subsystems[J]. Data and Knowledge Engineering Journal, 2006,58(2): 107-128.
[17]	TANENBAUM A S . Modern Operating Systems[M]. Second Ed Prentice Hall,Upper Saddle River,NJ, 2001.
[18]	LYER S , DRUSCHEL P . Anticipatory scheduling:a disk scheduling framework to overcome deceptive idleness in synchronous I/O[A]. Proceedings of the 18th ACM Symposium on Operating Systems Principles[C]. New York,NY, 2001.
[19]	RUEMMLER C , WILKES J . An introduction to disk drive modeling[J]. IEEE Comput, 1994,27(3): 16-28.
[20]	IOzone file system benchmark[EB/OL]. .
[21]	KATCHER J . Postmark:a New File System Benchmark[R]. Technical Report TR 3022,Network Appliance, 1997.

调度算法	95%-Heuristic	AS	WPCAS
总预期数	384630	371464	361268
预期成功数	362706	364590	356694
预期失败数	21924	6874	4574
成功时延	335140 ms	214014 ms	140768 ms
失败时延	86536 ms	41244 ms	16250 ms
伪空闲周期长度	0.238 ms	0.113 ms	0.047 ms
吞吐量	27415 kbit/s	34231 kbit/s	45238 kbit/s