存储扩展流量控制缓存分配策略研究
|
摘要
SAN具有高性能、易扩展、高可靠等优点,已成为主流网络存储技术之一,广泛应用于各类数据中心,以满足企业日益增长的信息对大容量、高性能存储系统的需求。然而,数据丢失的风险随着存储容量的增加而增加,为了降低数据丢失风险、提高数据和应用系统的可用性,人门正在研究基于高速广域网的SAN互连,以构建基于存储广域网的远程数据镜像、备份和文件共享。
目前,常用的三种存储扩展技术是基于IP、基于同步光纤网和基于波分复用的存储扩展,从协议层次上看,前者采用IP协议承载FC协议、后者直接将FC协议通过光载波传输、而新一代的同步光纤网使用透明通用成帧规程GFP T (Transparent General Framing Procedure)将FC协议映射到SONET中传输。与一般网络应用相比,存储扩展的特殊性主要有两点,其一是数据传输量大,需要网络具有高吞吐率;其二是可靠性要求高。流量控制是实现存储扩展应用中平滑数据流量、提高网络吞吐力和数据传输可靠性的重要技术手段。
研究表明,存储扩展应用中的IP网络具有高带宽时延积特性,该环境下,传统TCP协议受其窗口字段数据位长度的限制不能充分利用网络带宽,必须采用高速TCP协议。运用理论分析和仿真相结合的方法,研究了高速TCP协议对高带宽时延积网络的适应性,包括路由器缓存的需求、协议的公平性以及光突发交换网络对TCP协议的影响等内容。研究结果表明:(1)路由器的缓存容量只要不低于网络带宽时延积的10%,扩展链路带宽的利用率就能达到90%以上;(2)常见的四种高速TCP协议中,公平性由高到低的次序是FAST TCP、HSTCP、BIC TCP、STCP。采用自动队列管理能有效降低同步丢包的概率,提高TCP协议的公平性;(3)虽然光突发交换网的偏置时间和封装时间会对存储扩展性能产生一定的影响,但光网络的高稳定性和低丢包率有利于性能的提高,因此,采用光网络对IP存储扩展性能影响不大。上述研究结果表明,基于高速TCP协议的IP存储扩展可以在现有IP网络构架下实施,研究结果也为选择具体的TCP协议,实施基于IP的存储扩展提供了依据。
不同于IP网络,基于波分复用和基于SONET的存储扩展采用的是基于信用(Credit)的流量控制协议。研究表明,传统FC流量控制协议采用固定分配Credit的方法,导致存储扩展性能低、连路利用率不高。针对上述不足,提出了一种C23+模式的改进算法,新算法根据扩展链路的距离和带宽分配Credit,并将FC协议中的相对Credit更新策略修改成绝对更新策略,对的更新周期也进行了调整,理论分析表明,修改后的流量控制协议不仅能消除Credit更新信息丢失的累积效应,而且还能有效控制缓冲区上溢、缓冲区下溢、Credit下溢。Credit
在采用GFP T的SONET存储扩展应用中,GFP T网关和附接的FC设备之间采用的是基于Credit的流量控制协议,而两个GFP T网关之间使用的是简单流量控制协议ASFC(Alternate Simple Flow Control),分析发现,ASFC流量控制协议存在下列不足:(1)不能控制目标GFP T网关缓冲区的溢出;(2)流量控制原语的触发条件没有与目标GFP T网关设备的拥塞相关联。针对上述不足,提出了一种优化方法,使GFP T网关缓存容量的分配与链路距离及带宽相关联,同时还使ASFC流控原语的触发条件与FC设备的数据拥塞和目标GFP T网关数据拥塞关联。
在上述研究的基础上,构建了基于粗波分复用的存储扩展原型系统,并对远程访问性能进行了测试。
SAN(Storage Area Network) is the main network storage technology and has been widely used in the data center to meet the incremental requirements of the enterprise storage system with the large capacity, high performance. However, the risk of losing data will increase with the augumentation of storage capacity. In order to migrate the risk of losing data and improve the availability of data and application system, the connection of SAN based on high speed WAN for the aims of remote mirror, backup and file sharing has been researched.
At present, the main techniques for storage extension are based on IP, SONET and CWDM respectively. From the perspective of the protocol level, the first method carry the FC(Fibre channel) protocol with the IP protocol, the new generation SONET maps the FC protocol into the SONET by transparent GFP(Generic Framing Procedure), and meanwhile the CWDM technique carry the FC protocol by optical network directly.
Comparison with the conventional network application, the two special aspects the storage extension application are mainly expressed as follow: for one thing, it is necessary for the network to hold high throughput for the large amount of the data transmission. For another, the reliability is indispensable. Flow control is the important technique to smooth the data flow in network, enhance the network throughput and improve the reliability of the data transmission in the application of the storage extension. Research indicates that the IP networks for the storage extension have the character of high band latency production. Under such application background, the traditional TCP protocol can not make fully use of the network bandwidth for the constraint of data bit length of the window field, the high speed TCP protocol must be taken into consideration. For the high speed TCP protocol, the requirement of the router buffer, the fairness of the protocol and the influence of the optical switch network on the TCP protocol are studied by the methods of theoretical analyses and simulation. The study shows us that: (1) If the router buffer memory capacity is not lower than the 10% that of the band latency production, the utilization of the link bandwidth will achieve over 90%. (2) The sortorder of high speed TCP protocol fairness from high to low is FAST TCP, HSTCP, BIC TCP and standard TCP. The fairness of the TCP protocol can be improved through reducing the packet loss by using automatic queue management. (3)Although the offset time and assembly time of OBS(Optical Burst Switch) would have some effects on the storage extension performance, high stability and low packet loss of the OBS will help to improve the performance, therefore, the application of the optical network will have little influence on the storage extension performance. The above studies indicate that the IP storage extension with high speed TCP protocol can be deployed in current IP network infrastructure, and it also provide the guide for storage extension based on IP network.
Different from IP network, the flow control protocol of the storage extension based on WDM and SONET are based on credit with static allocation strategy, which leads to the low performance and low utilization ratio of network. An new algorithm of credit allocation named C23+ has been put forward for storage extension based on SONET and WDM. In the new algorithm the Credit is allocated according to the distance and bandwidth of the extension link, the absolute update policy is replaced with the relative update policy in the FC protocol, the update cycle is also adjusted. Theoretical analyses result show us that the new flow control protocol can not only migrate the accumulation effects from the information lose of the Credit update, but also prevent the overflow of buffer and the underflow of buffer or Credit.
In the SONET storage extension based on GFP T, the transfer control protocol between the GFP T gateway and the connected FC devices is based on Credit, meanwhile, the ASFC(Alternate Simple Flow Control) protocol is applied to between the GFT T gateways. Study indicates that there are two disadvantages for ASFC protocol:(1)The overflow of the target GFP T gateway buffer can not be prevented by ASFC flow control protocol. (2)The trigger condition of the flow control Primitive Signals are not related with the congest of the target GFT T gateway devices. A optimal method is presented, it allocates the credits for GFP T gateway according to distance and bandwidth of storage extension link, and it amendment the trigger condition of the flow control Primitive Signals of ASFC according to the congestion of FC devices and target GFT T gateway devices.
Based on the above research, a storage extension prototype based on CWDM is implemented and the test is done for the evaluation the performance of the remote access.
目前,常用的三种存储扩展技术是基于IP、基于同步光纤网和基于波分复用的存储扩展,从协议层次上看,前者采用IP协议承载FC协议、后者直接将FC协议通过光载波传输、而新一代的同步光纤网使用透明通用成帧规程GFP T (Transparent General Framing Procedure)将FC协议映射到SONET中传输。与一般网络应用相比,存储扩展的特殊性主要有两点,其一是数据传输量大,需要网络具有高吞吐率;其二是可靠性要求高。流量控制是实现存储扩展应用中平滑数据流量、提高网络吞吐力和数据传输可靠性的重要技术手段。
研究表明,存储扩展应用中的IP网络具有高带宽时延积特性,该环境下,传统TCP协议受其窗口字段数据位长度的限制不能充分利用网络带宽,必须采用高速TCP协议。运用理论分析和仿真相结合的方法,研究了高速TCP协议对高带宽时延积网络的适应性,包括路由器缓存的需求、协议的公平性以及光突发交换网络对TCP协议的影响等内容。研究结果表明:(1)路由器的缓存容量只要不低于网络带宽时延积的10%,扩展链路带宽的利用率就能达到90%以上;(2)常见的四种高速TCP协议中,公平性由高到低的次序是FAST TCP、HSTCP、BIC TCP、STCP。采用自动队列管理能有效降低同步丢包的概率,提高TCP协议的公平性;(3)虽然光突发交换网的偏置时间和封装时间会对存储扩展性能产生一定的影响,但光网络的高稳定性和低丢包率有利于性能的提高,因此,采用光网络对IP存储扩展性能影响不大。上述研究结果表明,基于高速TCP协议的IP存储扩展可以在现有IP网络构架下实施,研究结果也为选择具体的TCP协议,实施基于IP的存储扩展提供了依据。
不同于IP网络,基于波分复用和基于SONET的存储扩展采用的是基于信用(Credit)的流量控制协议。研究表明,传统FC流量控制协议采用固定分配Credit的方法,导致存储扩展性能低、连路利用率不高。针对上述不足,提出了一种C23+模式的改进算法,新算法根据扩展链路的距离和带宽分配Credit,并将FC协议中的相对Credit更新策略修改成绝对更新策略,对的更新周期也进行了调整,理论分析表明,修改后的流量控制协议不仅能消除Credit更新信息丢失的累积效应,而且还能有效控制缓冲区上溢、缓冲区下溢、Credit下溢。Credit
在采用GFP T的SONET存储扩展应用中,GFP T网关和附接的FC设备之间采用的是基于Credit的流量控制协议,而两个GFP T网关之间使用的是简单流量控制协议ASFC(Alternate Simple Flow Control),分析发现,ASFC流量控制协议存在下列不足:(1)不能控制目标GFP T网关缓冲区的溢出;(2)流量控制原语的触发条件没有与目标GFP T网关设备的拥塞相关联。针对上述不足,提出了一种优化方法,使GFP T网关缓存容量的分配与链路距离及带宽相关联,同时还使ASFC流控原语的触发条件与FC设备的数据拥塞和目标GFP T网关数据拥塞关联。
在上述研究的基础上,构建了基于粗波分复用的存储扩展原型系统,并对远程访问性能进行了测试。
SAN(Storage Area Network) is the main network storage technology and has been widely used in the data center to meet the incremental requirements of the enterprise storage system with the large capacity, high performance. However, the risk of losing data will increase with the augumentation of storage capacity. In order to migrate the risk of losing data and improve the availability of data and application system, the connection of SAN based on high speed WAN for the aims of remote mirror, backup and file sharing has been researched.
At present, the main techniques for storage extension are based on IP, SONET and CWDM respectively. From the perspective of the protocol level, the first method carry the FC(Fibre channel) protocol with the IP protocol, the new generation SONET maps the FC protocol into the SONET by transparent GFP(Generic Framing Procedure), and meanwhile the CWDM technique carry the FC protocol by optical network directly.
Comparison with the conventional network application, the two special aspects the storage extension application are mainly expressed as follow: for one thing, it is necessary for the network to hold high throughput for the large amount of the data transmission. For another, the reliability is indispensable. Flow control is the important technique to smooth the data flow in network, enhance the network throughput and improve the reliability of the data transmission in the application of the storage extension. Research indicates that the IP networks for the storage extension have the character of high band latency production. Under such application background, the traditional TCP protocol can not make fully use of the network bandwidth for the constraint of data bit length of the window field, the high speed TCP protocol must be taken into consideration. For the high speed TCP protocol, the requirement of the router buffer, the fairness of the protocol and the influence of the optical switch network on the TCP protocol are studied by the methods of theoretical analyses and simulation. The study shows us that: (1) If the router buffer memory capacity is not lower than the 10% that of the band latency production, the utilization of the link bandwidth will achieve over 90%. (2) The sortorder of high speed TCP protocol fairness from high to low is FAST TCP, HSTCP, BIC TCP and standard TCP. The fairness of the TCP protocol can be improved through reducing the packet loss by using automatic queue management. (3)Although the offset time and assembly time of OBS(Optical Burst Switch) would have some effects on the storage extension performance, high stability and low packet loss of the OBS will help to improve the performance, therefore, the application of the optical network will have little influence on the storage extension performance. The above studies indicate that the IP storage extension with high speed TCP protocol can be deployed in current IP network infrastructure, and it also provide the guide for storage extension based on IP network.
Different from IP network, the flow control protocol of the storage extension based on WDM and SONET are based on credit with static allocation strategy, which leads to the low performance and low utilization ratio of network. An new algorithm of credit allocation named C23+ has been put forward for storage extension based on SONET and WDM. In the new algorithm the Credit is allocated according to the distance and bandwidth of the extension link, the absolute update policy is replaced with the relative update policy in the FC protocol, the update cycle is also adjusted. Theoretical analyses result show us that the new flow control protocol can not only migrate the accumulation effects from the information lose of the Credit update, but also prevent the overflow of buffer and the underflow of buffer or Credit.
In the SONET storage extension based on GFP T, the transfer control protocol between the GFP T gateway and the connected FC devices is based on Credit, meanwhile, the ASFC(Alternate Simple Flow Control) protocol is applied to between the GFT T gateways. Study indicates that there are two disadvantages for ASFC protocol:(1)The overflow of the target GFP T gateway buffer can not be prevented by ASFC flow control protocol. (2)The trigger condition of the flow control Primitive Signals are not related with the congest of the target GFT T gateway devices. A optimal method is presented, it allocates the credits for GFP T gateway according to distance and bandwidth of storage extension link, and it amendment the trigger condition of the flow control Primitive Signals of ASFC according to the congestion of FC devices and target GFT T gateway devices.
Based on the above research, a storage extension prototype based on CWDM is implemented and the test is done for the evaluation the performance of the remote access.
引文
[1] Qiu Xiangqun, Telikepalli, R., Drwiega, T., et al. Reliability and availability assessment of storage area network extension solutions. IEEE Communications Magazine, 2005, 3(48): 80-85
[2]邓劲生，李宝峰，李蕾等．存储区域网络设计--现光纤通道和lP SAN的实用指南(第2版)．北京：电子工业出版社，2005：210～238
[3] G Gibson, R Meter. Network attached storage architecture. Communications of theACM, 2000, 43(11): 37-45
[4] Wang, P., Gilligan, R. E., Green, H et al. IP SAN - from iSCSI to IP-addressableEthernet disks. In: 20th IEEE/1 lth NASA Goddard Conference on Mass StorageSystems and Technologies (MSST 2003). San Diego, CA, United States: Institute ofElectrical and Electronics Engineers Inc, 2003: 189-193
[5]陈凯，白英彩．网络存储技术及发展趋势．电子学报，2002，30(12A)：1928～1932
[6]杨雨春，余胜生，周敬利．光纤通道适配器基本原理及其实现的研究．现代计算机，2001，29(7)10～13
[7]周敬利，韦丰，林芝等．基于光纤通道仲裁环的存储区域网的分析与实现．小型微型计算机系统，2004，25(1)：16～20
[8] Jaeyong Koh, Taewoong Kim, and Heonshik Shin. Scheduling real time messages infibre channel arbitrated loop. IFAC Control Engineering Practice, 1998, 6(1):119-127
[9] Alan F. Benner. Fibre channel for SANs.胡先志,胡佳妮.北京:人民邮电出版社,2003: 124-132
[10] Heywood, Mark. Keeping your data available in 10 steps. Network Security, 2005,2005(3): 14-16
[11]田均，黄利荣．远程容灾备份实现与原理．计算机工程与设计，2006，27(6)：1005～1007
[12]杨晓红，李健，杨卫国．信息系统容灾技术的分析与研究．计算机工程与设计，2005，26(10)：2727～2729
[13]杨朝红，宫云战．基于主从异步复制技术的容灾实时系统的研究与实现．计算机研究与发展，2003，40(7)：1105～1019
[14] C Pu, ALeff. Replica control in distributed systems: An asynchronous approach. In:Proc of the 1991 ACM SIGMOD Int'l Conf on Management of Data. Denver,Colorado: ACM Press, 1991: 377-386
[15] Ou YangKai, Zhou Jing-li, Guo Hui et al. NDMP-Plus: An Improved Protocol forNetwork Based Data Management. Journal of Shanghai University, 2005, 9(6):510-516
[16] C. M. Gauger, M. Kohn, S. Gunreben et al. Modeling and performance Evaluation ofiSCSI Storage Area Network over TCP/IP-based MAN and WAN network. In: 2ndInternational Conference on Broadband Networks. Boston: Institute of Electrical andElectronics Engineers Computer Society, Piscataway, NJ 08855-1331, United States,2005: 915-923
[17]朱茜．城域网也要SAN．中国计算机报，2004．3．1
[18] Qiu, Xiangqun, Telikepalli, Radha et al. Reliability and availability assessment ofstorage area network extension solutions. IEEE Communications Magazine, 2005,43(3): 80-85
[19] Fu, Xianglin, Liu, Zhaobin; Cao, Qiang. Study on the network protocol of the IP-based storage area network.. In: Sixth International Symposium on Optical Storage(ISOS 2002). Wuhan, China: The International Society for Optical Engineering, 2003:362-365
[20] Anon. Data trafficking. Electronic Product Design, 2004, 25(2): 45-46
[21] Gorshe, Steven S., Wilson, Trevor. Transparent Generic Framing Procedure (GFP): Aprotocol for efficient transport of block-coded data through SONET/SDH networks.IEEE Communications Magazine, 2002, 40(5): 88-95
[22]舒继武，薛巍，李必刚等．一种高可扩展存储网络系统TH—MSNS的研究与实现．计算机学报，2005，3(28)：326～333
[23] BI Jijun, SUN Xiaohan, ZHANG Mingde. et al. Studies on the key technologies ofCWDM. Chinese Journal of Electron Device, 2002, 25(4): 369-373
[24] Scholten, M., Zhenyu Zhu; Hernandez-Valencia, E et al. Data transport applicationsusing GFP IEEE Communications Magazine, 2002, 40(5): 96-103
[25] Hernandez-Valencia, E, Michael Scholten, Zhenyu Zhu. The Generic FramingProcedure (GFP): An Overview. IEEE Communications Magazine, 2002, 40(5):63-71
[26]苏文，王忠民．IP存储技术研究．计算机应用研究，2004，6：240～242
[27]魏轶伟，熊剑平，贾惠波．基于IP的存储网络技术．计算机工程与应用，2002，13：156～158
[28] FCIP: Tom Clark. Designing Storage Area Networks: A Practical Reference for Implementing Storage Area Networks. Boston: Addi son-Wesley Longman Publishing Co., Inc: 150-323
[29]傅湘林，韩德志，谢长生．基于IP的SAN互联技术的研究．计算机科学，2006，30(6)：46～49
[30] Siegel Eric, Terrill Bill. Extending the SAN over the WAN. IT Architect, 2005,20(11): 14-16
[31] Gorshe, Steven S, Wilson, et al. Transparent Generic Framing Procedure (GFP): Aprotocol for efficient transport of block-coded data through SONET/SDH networks.IEEE Communications Magazine, 2002, 40(5): 88-95
[32] Scholten Mike, Zhu Zhenyu, Hernandez-Valencia Enrique et al. Data transportapplications using GFP. IEEE Communications Magazine, 2002, 40(5): 96-103
[33] Armstrong Tim, Gorshe Steven S. Generic framing procedure (GFP) and data overSONET/SDH and OTN. IEEE Communications Magazine, 2002, 40(5): 60-61
[34]孙嵘．通用成帧规程GFP协议结构特点及其关键技术．光通信技术，2003，3：17～23
[35] Pranggono Bernardi, Elmirghani Jaafar M. H. A novel optical storage area network implemented in a metro WDM setting. In 2005 Workshop on High Performance Switching and Routing, HPSR 2005. Hong Kong, China: Institute of Electrical and Electronics Engineers Computer Society, Piscataway, NJ 08855-1331, United States, 2005: 49-52
[36] Rope, Todd. Pluggables make WDM more cost-effective. Lightwave, 2004, 21(4): 18-21
[37] El-Gorashi Taisir E. H, Pranggono Bernardi, Elmirghani Jaafar M. H. Multi-wavelength metro WDM sectioned ring for SAN extension under hot node scenario and variable traffic profiles. In: 2006 International Conference on Transparent Optical Networks, ICTON 2006. Nottingham, United Kingdom: Institute of Electrical and Electronics Engineers Computer Society, Piscataway, NJ 08855-1331, United States, 2006: 139-146
[38]罗万明，林闯，阎保平．TCP／IP拥塞控制研究．计算机学报，2001，24(1)：1～18
[39] PadhyeJ, Firoiu V, Towsley D. et al. Modeling TCP throughput: A simple model and its empirical validation IEEE/ ACM Transactions on Networking, 2000, 8(2): 133-145
[40] C. Casetti, M. Gerla, S. Mascolo, M. et al. TCP Westwood: end-to-end congestioncontrol for wired/wireless networks. Wireless Netw, 2002., 8: 467-479
[41] S. Floyd. High Speed TCP for large congestion windows. IEFT RFC 3649, December2003
[42] T. Kelly. Scalable TCP: improving performance in high speed wide area networks.Computer Communication, 2003, 32(2): 83-91
[43] Xu L, Harfoush K, Rhee I. Binary increase congestion control for fast long distancenetworks. In: Proceedings of the IEEE INFOCOM. Hong Kong, 2004: 2514-2524
[44] Jin C, Wei D. X., Low S. H.. FAST TCP: Motivation, architecture, algorithms,performance. In: Proceedings of the IEEE INFOCOM, Hong Kong, 2004: 2490—2501
[45] Curtis Villamizar and Cheng Song. High performance TCP in ANSNET. ACMcomputer Communications Review, 2004, 24(5): 45-60
[46] Mascolo, Saverio, Vacirca, Francesco. Congestion control and sizing router buffers inthe internet. In: Proceedings of the 44th IEEE Conference on Decision and Control,and the European Control Conference, CDC-ECC '05. Seville, Spain: Institute ofElectrical and Electronics Engineers Computer Society, Piscataway, NJ 08855-1331,United States, 2005. 6750-6755
[47] Dhamdhere, A. Dovrolis, C. Open issues in router buffer sizing. ComputerCommunication Review, 2006, 36(1): 87-92
[48] Raina, G. Towsley, D., Wischik, D. Part II: control theory for buffer sizing. ComputerCommunication Review, 2005, 35(3): 79-82
[49] Ganjali, Y. McKeown, N. Update on buffer sizing in Internet routers. ComputerCommunication Review, 2006, 36(5): 67-70
[50] Appenzeller, G. Keslassy, I, McKeown, N. Sizing router buffers. ComputerCommunication Review, 2004, 34(4): 281-292
[51] Li-song Shao, He-ying Zhang, Wen-hua Dou. General window-based congestioncontrol: buffer occupancy, network efficiency and packet loss. In: Proceedings of the4th Annual Communication Networks and Services Research Conference. LosAlamitos: IEEE Computer Society, 2006. 7-15
[52] Frank Kelly, S. Zachary, I. Ziedins. Notes on Effective Bandwidths. Stochasticnetwork: Theory and Applications of Royal Statistical Society Lecture NotesSeries(4). Oxford: Oxford University Press, 1996. 141-168
[53] Barman Dhiman; Smaragdakis Georgios, Matta Ibrahim. The effect of router buffersize on highspeed TCP performance. In: Proceedings of GLOBECOM'04 - IEEEGlobal Telecommunications Conference. Dallas, TX, United States: Institute ofElectrical and Electronics Engineers Inc., New York, NY 10016-5997, United States,2004. 1617-1621
[54] K. Avrachenkov, U. Ayesta, E. Altman et al. The Effect of Router Buffer Size on theTCP Performance. In: Proceedings of LONIIS workshop on TelecommunicationNetworks and TeletrafficTheory, StPetersburg, Russia, 2002. 116-121
[55] David McDysan. QoS & Traffic Management in IP & ATM Networks. New York:McGrawHill companies, 2000. 162-167
[56] Thomas Bonald, Laurent Massoulie. Impact of fairness on Internet performance.ACM SIGMETRICS Performance Evaluation Review, 2001, 29(1): 82-91
[57] Massoulie L, Roberts J. Bandwidth Sharing: Objectives and Fairness. ProceedingsIEEEINFOCOM, 1999(3): 1395-1403
[58]张敬辕，谢剑英，王明中．Internet带宽分配的公平性研究．计算机工程，2002，28(3)：154～156
[59]Long Chennian, Yang Huilong, Liu Zhixin et al. Fairness performance analysis:HSTCP/Drop Tairl vs HSTCP/AQM. In: Proceedings of the 25th Chinese ControlConference. Harbin, Heilongjiang: IEEE, 2006. 1807-1810
[60]S. H. Low. A Duality Model of TCP and Queue Management Algorithms. IEEE/ACMTransactions on Networking, 200311(4): 525-536
[61] PLiansheng Tan, Lina Dong, Cao Yuan et al. Fairness comparison of FAST TCP andTCP Reno. Computer Communications, 2007, 30(6): 1375-1382
[62]Damien Phillips, Jiankun Hu. Enhancements for RTT-Fair Highspeed TCP. In:Proceedings of Global Telecommunications Conference, 2006. GLOBECOM '06. SanFrancisco: IEEE, 2006. 1-5
[63] Lisong Xu, Khaled Harfoush, Injong Rhee. Binary Increase Congestion Control forFast, Long Distance Networks. In: Proceedings of INFOCOM 2004-Twenty-thirdAnnual Joint Conference of the IEEE Computer and Communications Societies.North Carolina State: IEEE, 2005. 2514-2524
[64] S. Bohace, J. P. Hespanha, J. LEE, and K. ObRACZKA. A hybrid systems modelingframework for fast and accurate simulation of data communication networks, ACMSIGMETRICS Performance Evaluation Review, 2003, 31(1): 58-69
[65] L. Zhang, D. D. Clark. Oscillating behavior of network traffic: A case studysimulation. Internet working: Research and Experience, 1999,1(1): 101-112
[66] L. Qiu, Y Zhang, S. Keshav. Understanding the performance of many TCP flows.Computer Networks, 2001, 37(2): 277-306
[67] M. Christiansen, K. Jaffay, D. OTT. Tuning RED for web traffic. Networking,IEEE/ACM Transactions on, 2001, 9(3): 249-264
[68] Qiao C M, Yoo M. Optical Burst Switching-a new Paradigm for an Optical Internet.Journal of High Speed Networks, 1999, 8(1): 69-84
[69]潘永康，王晟，张梭．TCP over OBS的性能分析及改进．计算机应用，2007，27(4)：808-810
[70] Yu X, Li J K, Cao X J, et al. Traffic Statistics and Performance Evaluation in Optical Burst Switched Networks. Journal of Light wave Technology, 2004, 22(12): 2722-2738
[71]文爱军，靖杰，刘建平．光突发交换网特性对TCP性能的影响．西安电子科技大学学报，2006，33(5)：748-753
[72]刘建平，刘增基，文爱军．OBS网络中汇聚机制对边缘节点时延特性的影响．计算机科学，2006，33(2)：52-54
[73] Andrea D, Marco L. Impact of Segments Aggregation on TCP Reno Flows in Optical Burst Switching Network. In: Proceedings of the INFOCOM'2002. New York: IEEE INFOCOM, 2002. 1803-1812
[74]焦硕，吴海珊，徐坤等．TCP在OBS网络环境中的性能的建模分析．北京邮电大学学报，2005，28(6)：39-42
[75] Padhye J, Firoiu V, Towsley D F, et al. Modeling TCP reno performance: a simplemodel and its empirical validation. IEEE/ACM Trans Networking, 2000, 8(2):133-145
[76] Qin Leihua, Yu shengsheng, Zhou jingli. Analysis and amendment of flow controlcredit-based in SAN extension. In: Proceedings of SPIE vol. 6022. Shanghai: TheInternational Society for Optical Engineering, 2005. 869-876
[77] ALLounsuruy. Storage over SONET/SDH connectivity. Computer Technology, 2003,23(5): 25-30
[78] Douglas W Browning. Flow Control in High-Speed Communication Networks. IEEETransactions, 1994, 42(7): 2480-2489
[79] Schoenen R, Post G, Muller A. Analysis and dimensioning of credit-based flowcontrol for the ABR service in ATM networks. In: Proceedings of GlobalTelecommunications Conference, 1998. GLOBECOM 98. Sydney, NSW: IEEE, 1998.2399-2404
[80] Kung H. T, Koling Chang. Receiver-oriented adaptive buffer allocation in credit-based flow control for ATM networks. In: Proceedings of INFOCOM '95. Fourteenth Annual Joint Conference of the IEEE Computer and Communications Societies. Boston, MA: IEEE, 1995. 239-252
[81]冯立宏，赵林亮，王光兴．ATM网络流量控制中的一种信元丢弃机制．计算机研究与发展，2000，37(11)：1340-1343
[82] Schoenen R, Dahlhoff A. Closed loop credit-based flow control with internal backpressure in input and output queued switches. In: Proceedings of High Performance Switching and Routing, 2000. ATM 2000. Heidelber: IEEE, 2000. 195-203
[83] H. T. Kung, A. Chapman. The FCVC (Flow-Controlled Virtual Channels) Proposal for ATM Networks: A summary. In: Proceedings of International Conf. on Network Protocols, 1993. San Francisco, California: IEEE, 1993. 116-127
[84] H. T. Kung, Rober Morris Thomas Charuhas et al. Use of Link by Link Flow control in Maximizing ATM Network Performance: Simulation Results. In: Proceedings of IEEE Hot Interconnects Symposium. Palo Alto, California: IEEE, 1993: 1-12
[85] H. T. Kung, Trevor Blackwell, Alan Chapman. Credit-based flow control for ATM networks: credit update protocol, adaptive credit allocation and statistical multiplexing. ACM SIGCOMM Computer Communication Review, 1994, 24(4): 101-114
[86]吴丽征．基于SONET／SDH的存储网络容灾技术分析．微电子学与计算机，2006，23(6)：119～122
[87]周敬利，马吉峰，秦磊华．应用于远程异地存储FC—SAN的CWDM实现方法．计算机工程与科学，2006，28(1)：93-100
[88] Y Lu, D H C Du. Performance study of iscsi-based storage subsystems. IEEECommunications Magazine, 2003, 41(3): 76-82
[89] B Cui, J Liu, G wang et al. Study on Performance of IP-SWAN Based on DistributedNS-RAID. In: Proceedings of 28th Annual International Conference on ComputerSoftware and Applications. Hong Kong: IEEE Computer Society, 2004. 566-571
[90] Radha Telikepalli, Tadeusz Drwiega, James Yan et al. Storage Area NetworkExtension Solutions and Their Performance Assessment. IEEE CommunicationsMagazine, 2004, 42(4): 56-63
[2]邓劲生，李宝峰，李蕾等．存储区域网络设计--现光纤通道和lP SAN的实用指南(第2版)．北京：电子工业出版社，2005：210～238
[3] G Gibson, R Meter. Network attached storage architecture. Communications of theACM, 2000, 43(11): 37-45
[4] Wang, P., Gilligan, R. E., Green, H et al. IP SAN - from iSCSI to IP-addressableEthernet disks. In: 20th IEEE/1 lth NASA Goddard Conference on Mass StorageSystems and Technologies (MSST 2003). San Diego, CA, United States: Institute ofElectrical and Electronics Engineers Inc, 2003: 189-193
[5]陈凯，白英彩．网络存储技术及发展趋势．电子学报，2002，30(12A)：1928～1932
[6]杨雨春，余胜生，周敬利．光纤通道适配器基本原理及其实现的研究．现代计算机，2001，29(7)10～13
[7]周敬利，韦丰，林芝等．基于光纤通道仲裁环的存储区域网的分析与实现．小型微型计算机系统，2004，25(1)：16～20
[8] Jaeyong Koh, Taewoong Kim, and Heonshik Shin. Scheduling real time messages infibre channel arbitrated loop. IFAC Control Engineering Practice, 1998, 6(1):119-127
[9] Alan F. Benner. Fibre channel for SANs.胡先志,胡佳妮.北京:人民邮电出版社,2003: 124-132
[10] Heywood, Mark. Keeping your data available in 10 steps. Network Security, 2005,2005(3): 14-16
[11]田均，黄利荣．远程容灾备份实现与原理．计算机工程与设计，2006，27(6)：1005～1007
[12]杨晓红，李健，杨卫国．信息系统容灾技术的分析与研究．计算机工程与设计，2005，26(10)：2727～2729
[13]杨朝红，宫云战．基于主从异步复制技术的容灾实时系统的研究与实现．计算机研究与发展，2003，40(7)：1105～1019
[14] C Pu, ALeff. Replica control in distributed systems: An asynchronous approach. In:Proc of the 1991 ACM SIGMOD Int'l Conf on Management of Data. Denver,Colorado: ACM Press, 1991: 377-386
[15] Ou YangKai, Zhou Jing-li, Guo Hui et al. NDMP-Plus: An Improved Protocol forNetwork Based Data Management. Journal of Shanghai University, 2005, 9(6):510-516
[16] C. M. Gauger, M. Kohn, S. Gunreben et al. Modeling and performance Evaluation ofiSCSI Storage Area Network over TCP/IP-based MAN and WAN network. In: 2ndInternational Conference on Broadband Networks. Boston: Institute of Electrical andElectronics Engineers Computer Society, Piscataway, NJ 08855-1331, United States,2005: 915-923
[17]朱茜．城域网也要SAN．中国计算机报，2004．3．1
[18] Qiu, Xiangqun, Telikepalli, Radha et al. Reliability and availability assessment ofstorage area network extension solutions. IEEE Communications Magazine, 2005,43(3): 80-85
[19] Fu, Xianglin, Liu, Zhaobin; Cao, Qiang. Study on the network protocol of the IP-based storage area network.. In: Sixth International Symposium on Optical Storage(ISOS 2002). Wuhan, China: The International Society for Optical Engineering, 2003:362-365
[20] Anon. Data trafficking. Electronic Product Design, 2004, 25(2): 45-46
[21] Gorshe, Steven S., Wilson, Trevor. Transparent Generic Framing Procedure (GFP): Aprotocol for efficient transport of block-coded data through SONET/SDH networks.IEEE Communications Magazine, 2002, 40(5): 88-95
[22]舒继武，薛巍，李必刚等．一种高可扩展存储网络系统TH—MSNS的研究与实现．计算机学报，2005，3(28)：326～333
[23] BI Jijun, SUN Xiaohan, ZHANG Mingde. et al. Studies on the key technologies ofCWDM. Chinese Journal of Electron Device, 2002, 25(4): 369-373
[24] Scholten, M., Zhenyu Zhu; Hernandez-Valencia, E et al. Data transport applicationsusing GFP IEEE Communications Magazine, 2002, 40(5): 96-103
[25] Hernandez-Valencia, E, Michael Scholten, Zhenyu Zhu. The Generic FramingProcedure (GFP): An Overview. IEEE Communications Magazine, 2002, 40(5):63-71
[26]苏文，王忠民．IP存储技术研究．计算机应用研究，2004，6：240～242
[27]魏轶伟，熊剑平，贾惠波．基于IP的存储网络技术．计算机工程与应用，2002，13：156～158
[28] FCIP: Tom Clark. Designing Storage Area Networks: A Practical Reference for Implementing Storage Area Networks. Boston: Addi son-Wesley Longman Publishing Co., Inc: 150-323
[29]傅湘林，韩德志，谢长生．基于IP的SAN互联技术的研究．计算机科学，2006，30(6)：46～49
[30] Siegel Eric, Terrill Bill. Extending the SAN over the WAN. IT Architect, 2005,20(11): 14-16
[31] Gorshe, Steven S, Wilson, et al. Transparent Generic Framing Procedure (GFP): Aprotocol for efficient transport of block-coded data through SONET/SDH networks.IEEE Communications Magazine, 2002, 40(5): 88-95
[32] Scholten Mike, Zhu Zhenyu, Hernandez-Valencia Enrique et al. Data transportapplications using GFP. IEEE Communications Magazine, 2002, 40(5): 96-103
[33] Armstrong Tim, Gorshe Steven S. Generic framing procedure (GFP) and data overSONET/SDH and OTN. IEEE Communications Magazine, 2002, 40(5): 60-61
[34]孙嵘．通用成帧规程GFP协议结构特点及其关键技术．光通信技术，2003，3：17～23
[35] Pranggono Bernardi, Elmirghani Jaafar M. H. A novel optical storage area network implemented in a metro WDM setting. In 2005 Workshop on High Performance Switching and Routing, HPSR 2005. Hong Kong, China: Institute of Electrical and Electronics Engineers Computer Society, Piscataway, NJ 08855-1331, United States, 2005: 49-52
[36] Rope, Todd. Pluggables make WDM more cost-effective. Lightwave, 2004, 21(4): 18-21
[37] El-Gorashi Taisir E. H, Pranggono Bernardi, Elmirghani Jaafar M. H. Multi-wavelength metro WDM sectioned ring for SAN extension under hot node scenario and variable traffic profiles. In: 2006 International Conference on Transparent Optical Networks, ICTON 2006. Nottingham, United Kingdom: Institute of Electrical and Electronics Engineers Computer Society, Piscataway, NJ 08855-1331, United States, 2006: 139-146
[38]罗万明，林闯，阎保平．TCP／IP拥塞控制研究．计算机学报，2001，24(1)：1～18
[39] PadhyeJ, Firoiu V, Towsley D. et al. Modeling TCP throughput: A simple model and its empirical validation IEEE/ ACM Transactions on Networking, 2000, 8(2): 133-145
[40] C. Casetti, M. Gerla, S. Mascolo, M. et al. TCP Westwood: end-to-end congestioncontrol for wired/wireless networks. Wireless Netw, 2002., 8: 467-479
[41] S. Floyd. High Speed TCP for large congestion windows. IEFT RFC 3649, December2003
[42] T. Kelly. Scalable TCP: improving performance in high speed wide area networks.Computer Communication, 2003, 32(2): 83-91
[43] Xu L, Harfoush K, Rhee I. Binary increase congestion control for fast long distancenetworks. In: Proceedings of the IEEE INFOCOM. Hong Kong, 2004: 2514-2524
[44] Jin C, Wei D. X., Low S. H.. FAST TCP: Motivation, architecture, algorithms,performance. In: Proceedings of the IEEE INFOCOM, Hong Kong, 2004: 2490—2501
[45] Curtis Villamizar and Cheng Song. High performance TCP in ANSNET. ACMcomputer Communications Review, 2004, 24(5): 45-60
[46] Mascolo, Saverio, Vacirca, Francesco. Congestion control and sizing router buffers inthe internet. In: Proceedings of the 44th IEEE Conference on Decision and Control,and the European Control Conference, CDC-ECC '05. Seville, Spain: Institute ofElectrical and Electronics Engineers Computer Society, Piscataway, NJ 08855-1331,United States, 2005. 6750-6755
[47] Dhamdhere, A. Dovrolis, C. Open issues in router buffer sizing. ComputerCommunication Review, 2006, 36(1): 87-92
[48] Raina, G. Towsley, D., Wischik, D. Part II: control theory for buffer sizing. ComputerCommunication Review, 2005, 35(3): 79-82
[49] Ganjali, Y. McKeown, N. Update on buffer sizing in Internet routers. ComputerCommunication Review, 2006, 36(5): 67-70
[50] Appenzeller, G. Keslassy, I, McKeown, N. Sizing router buffers. ComputerCommunication Review, 2004, 34(4): 281-292
[51] Li-song Shao, He-ying Zhang, Wen-hua Dou. General window-based congestioncontrol: buffer occupancy, network efficiency and packet loss. In: Proceedings of the4th Annual Communication Networks and Services Research Conference. LosAlamitos: IEEE Computer Society, 2006. 7-15
[52] Frank Kelly, S. Zachary, I. Ziedins. Notes on Effective Bandwidths. Stochasticnetwork: Theory and Applications of Royal Statistical Society Lecture NotesSeries(4). Oxford: Oxford University Press, 1996. 141-168
[53] Barman Dhiman; Smaragdakis Georgios, Matta Ibrahim. The effect of router buffersize on highspeed TCP performance. In: Proceedings of GLOBECOM'04 - IEEEGlobal Telecommunications Conference. Dallas, TX, United States: Institute ofElectrical and Electronics Engineers Inc., New York, NY 10016-5997, United States,2004. 1617-1621
[54] K. Avrachenkov, U. Ayesta, E. Altman et al. The Effect of Router Buffer Size on theTCP Performance. In: Proceedings of LONIIS workshop on TelecommunicationNetworks and TeletrafficTheory, StPetersburg, Russia, 2002. 116-121
[55] David McDysan. QoS & Traffic Management in IP & ATM Networks. New York:McGrawHill companies, 2000. 162-167
[56] Thomas Bonald, Laurent Massoulie. Impact of fairness on Internet performance.ACM SIGMETRICS Performance Evaluation Review, 2001, 29(1): 82-91
[57] Massoulie L, Roberts J. Bandwidth Sharing: Objectives and Fairness. ProceedingsIEEEINFOCOM, 1999(3): 1395-1403
[58]张敬辕，谢剑英，王明中．Internet带宽分配的公平性研究．计算机工程，2002，28(3)：154～156
[59]Long Chennian, Yang Huilong, Liu Zhixin et al. Fairness performance analysis:HSTCP/Drop Tairl vs HSTCP/AQM. In: Proceedings of the 25th Chinese ControlConference. Harbin, Heilongjiang: IEEE, 2006. 1807-1810
[60]S. H. Low. A Duality Model of TCP and Queue Management Algorithms. IEEE/ACMTransactions on Networking, 200311(4): 525-536
[61] PLiansheng Tan, Lina Dong, Cao Yuan et al. Fairness comparison of FAST TCP andTCP Reno. Computer Communications, 2007, 30(6): 1375-1382
[62]Damien Phillips, Jiankun Hu. Enhancements for RTT-Fair Highspeed TCP. In:Proceedings of Global Telecommunications Conference, 2006. GLOBECOM '06. SanFrancisco: IEEE, 2006. 1-5
[63] Lisong Xu, Khaled Harfoush, Injong Rhee. Binary Increase Congestion Control forFast, Long Distance Networks. In: Proceedings of INFOCOM 2004-Twenty-thirdAnnual Joint Conference of the IEEE Computer and Communications Societies.North Carolina State: IEEE, 2005. 2514-2524
[64] S. Bohace, J. P. Hespanha, J. LEE, and K. ObRACZKA. A hybrid systems modelingframework for fast and accurate simulation of data communication networks, ACMSIGMETRICS Performance Evaluation Review, 2003, 31(1): 58-69
[65] L. Zhang, D. D. Clark. Oscillating behavior of network traffic: A case studysimulation. Internet working: Research and Experience, 1999,1(1): 101-112
[66] L. Qiu, Y Zhang, S. Keshav. Understanding the performance of many TCP flows.Computer Networks, 2001, 37(2): 277-306
[67] M. Christiansen, K. Jaffay, D. OTT. Tuning RED for web traffic. Networking,IEEE/ACM Transactions on, 2001, 9(3): 249-264
[68] Qiao C M, Yoo M. Optical Burst Switching-a new Paradigm for an Optical Internet.Journal of High Speed Networks, 1999, 8(1): 69-84
[69]潘永康，王晟，张梭．TCP over OBS的性能分析及改进．计算机应用，2007，27(4)：808-810
[70] Yu X, Li J K, Cao X J, et al. Traffic Statistics and Performance Evaluation in Optical Burst Switched Networks. Journal of Light wave Technology, 2004, 22(12): 2722-2738
[71]文爱军，靖杰，刘建平．光突发交换网特性对TCP性能的影响．西安电子科技大学学报，2006，33(5)：748-753
[72]刘建平，刘增基，文爱军．OBS网络中汇聚机制对边缘节点时延特性的影响．计算机科学，2006，33(2)：52-54
[73] Andrea D, Marco L. Impact of Segments Aggregation on TCP Reno Flows in Optical Burst Switching Network. In: Proceedings of the INFOCOM'2002. New York: IEEE INFOCOM, 2002. 1803-1812
[74]焦硕，吴海珊，徐坤等．TCP在OBS网络环境中的性能的建模分析．北京邮电大学学报，2005，28(6)：39-42
[75] Padhye J, Firoiu V, Towsley D F, et al. Modeling TCP reno performance: a simplemodel and its empirical validation. IEEE/ACM Trans Networking, 2000, 8(2):133-145
[76] Qin Leihua, Yu shengsheng, Zhou jingli. Analysis and amendment of flow controlcredit-based in SAN extension. In: Proceedings of SPIE vol. 6022. Shanghai: TheInternational Society for Optical Engineering, 2005. 869-876
[77] ALLounsuruy. Storage over SONET/SDH connectivity. Computer Technology, 2003,23(5): 25-30
[78] Douglas W Browning. Flow Control in High-Speed Communication Networks. IEEETransactions, 1994, 42(7): 2480-2489
[79] Schoenen R, Post G, Muller A. Analysis and dimensioning of credit-based flowcontrol for the ABR service in ATM networks. In: Proceedings of GlobalTelecommunications Conference, 1998. GLOBECOM 98. Sydney, NSW: IEEE, 1998.2399-2404
[80] Kung H. T, Koling Chang. Receiver-oriented adaptive buffer allocation in credit-based flow control for ATM networks. In: Proceedings of INFOCOM '95. Fourteenth Annual Joint Conference of the IEEE Computer and Communications Societies. Boston, MA: IEEE, 1995. 239-252
[81]冯立宏，赵林亮，王光兴．ATM网络流量控制中的一种信元丢弃机制．计算机研究与发展，2000，37(11)：1340-1343
[82] Schoenen R, Dahlhoff A. Closed loop credit-based flow control with internal backpressure in input and output queued switches. In: Proceedings of High Performance Switching and Routing, 2000. ATM 2000. Heidelber: IEEE, 2000. 195-203
[83] H. T. Kung, A. Chapman. The FCVC (Flow-Controlled Virtual Channels) Proposal for ATM Networks: A summary. In: Proceedings of International Conf. on Network Protocols, 1993. San Francisco, California: IEEE, 1993. 116-127
[84] H. T. Kung, Rober Morris Thomas Charuhas et al. Use of Link by Link Flow control in Maximizing ATM Network Performance: Simulation Results. In: Proceedings of IEEE Hot Interconnects Symposium. Palo Alto, California: IEEE, 1993: 1-12
[85] H. T. Kung, Trevor Blackwell, Alan Chapman. Credit-based flow control for ATM networks: credit update protocol, adaptive credit allocation and statistical multiplexing. ACM SIGCOMM Computer Communication Review, 1994, 24(4): 101-114
[86]吴丽征．基于SONET／SDH的存储网络容灾技术分析．微电子学与计算机，2006，23(6)：119～122
[87]周敬利，马吉峰，秦磊华．应用于远程异地存储FC—SAN的CWDM实现方法．计算机工程与科学，2006，28(1)：93-100
[88] Y Lu, D H C Du. Performance study of iscsi-based storage subsystems. IEEECommunications Magazine, 2003, 41(3): 76-82
[89] B Cui, J Liu, G wang et al. Study on Performance of IP-SWAN Based on DistributedNS-RAID. In: Proceedings of 28th Annual International Conference on ComputerSoftware and Applications. Hong Kong: IEEE Computer Society, 2004. 566-571
[90] Radha Telikepalli, Tadeusz Drwiega, James Yan et al. Storage Area NetworkExtension Solutions and Their Performance Assessment. IEEE CommunicationsMagazine, 2004, 42(4): 56-63