zhouzg281@nenu.edu.cn,{xugc,dengcy}@jlu.edu.cn,hejx@jlu.edu.cn,jianhuajiang@yahoo.com Anycast Member Server B Anycast Member Server A



Similar documents
Quality of Service Routing Network and Performance Evaluation*

Tunnel Broker System Using IPv4 Anycast

A Topology-Aware Relay Lookup Scheme for P2P VoIP System

Network Layer Implemented Anycast Load Balancing

PERFORMANCE STUDY AND SIMULATION OF AN ANYCAST PROTOCOL FOR WIRELESS MOBILE AD HOC NETWORKS

Reliable Multicast Protocol with Packet Forwarding in Wireless Internet

Performance of networks containing both MaxNet and SumNet links

IP Anycast: Point to (Any) Point Communications. Draft 0.3. Chris Metz, Introduction

Internet Video Streaming and Cloud-based Multimedia Applications. Outline

Analysis of QoS Routing Approach and the starvation`s evaluation in LAN

Server-Based Dynamic Server Selection Algorithms

Multicast vs. P2P for content distribution

Load Balancing. Final Network Exam LSNAT. Sommaire. How works a "traditional" NAT? Un article de Le wiki des TPs RSM.

Optimizing Congestion in Peer-to-Peer File Sharing Based on Network Coding

packet retransmitting based on dynamic route table technology, as shown in fig. 2 and 3.

How To Monitor Performance On Eve

A ROUTING ALGORITHM FOR MPLS TRAFFIC ENGINEERING IN LEO SATELLITE CONSTELLATION NETWORK. Received September 2012; revised January 2013

CHAPTER 8 CONCLUSION AND FUTURE ENHANCEMENTS

A Catechistic Method for Traffic Pattern Discovery in MANET

Traceroute-Based Topology Inference without Network Coordinate Estimation

VoIP versus VoMPLS Performance Evaluation

Simulation of Heuristic Usage for Load Balancing In Routing Efficiency

Indirection. science can be solved by adding another level of indirection" -- Butler Lampson. "Every problem in computer

Explicit Multicast Routing

Efficient DNS based Load Balancing for Bursty Web Application Traffic

Top-Down Network Design

119, Munjiro, Yuseong-gu, Daejeon, Korea. {neofaith, mckim, torshong, 2 InfraLab, Korea Telecom

Definition. A Historical Example

Path Selection Analysis in MPLS Network Based on QoS

18: Enhanced Quality of Service

Computer Networks. A Top-Down Approach. Behrouz A. Forouzan. and. Firouz Mosharraf. \Connect Mc \ Learn. Hill

Content-Aware Load Balancing using Direct Routing for VOD Streaming Service

An Efficient Load Balancing Technology in CDN

QoSIP: A QoS Aware IP Routing Protocol for Multimedia Data

Content Delivery Networks. Shaxun Chen April 21, 2009

Datagram-based network layer: forwarding; routing. Additional function of VCbased network layer: call setup.

DESIGN OF CLUSTER OF SIP SERVER BY LOAD BALANCER

Fuzzy Active Queue Management for Assured Forwarding Traffic in Differentiated Services Network

A Study of Internet Packet Reordering

An Efficient QoS Routing Protocol for Mobile Ad-Hoc Networks *

3D On-chip Data Center Networks Using Circuit Switches and Packet Switches

Performance Analysis of AQM Schemes in Wired and Wireless Networks based on TCP flow

MAXIMIZING RESTORABLE THROUGHPUT IN MPLS NETWORKS

Final for ECE374 05/06/13 Solution!!

Design and Deployment of Locality-aware Overlay Multicast Protocol for Live Streaming Services

An Improved Available Bandwidth Measurement Algorithm based on Pathload

Network Security TCP/IP Refresher

Address Resolution Protocol (ARP)

Influence of Load Balancing on Quality of Real Time Data Transmission*

NQA Technology White Paper

Measurement of V2oIP over Wide Area Network between Countries Using Soft Phone and USB Phone

Performance Monitoring on Networked Virtual Environments

Transport layer issues in ad hoc wireless networks Dmitrij Lagutin,

Telecommunication Services Engineering (TSE) Lab. Chapter III 4G Long Term Evolution (LTE) and Evolved Packet Core (EPC)

CS101 Lecture 19: Internetworking. What You ll Learn Today

Using Fuzzy Logic Control to Provide Intelligent Traffic Management Service for High-Speed Networks ABSTRACT:

Faculty of Engineering Computer Engineering Department Islamic University of Gaza Network Chapter# 19 INTERNETWORK OPERATION

Transport and Network Layer

Management of Telecommunication Networks. Prof. Dr. Aleksandar Tsenov

10CS64: COMPUTER NETWORKS - II

Voice over IP. Overview. What is VoIP and how it works. Reduction of voice quality. Quality of Service for VoIP

The necessity of multicast for IPTV streaming

GLOBAL SERVER LOAD BALANCING WITH SERVERIRON

Integrating Internet Protocol (IP) Multicast over Multiprotocol Label Switching (MPLS) for Real Time Video Conferencing Data Transmission

International Journal of Advanced Research in Computer Science and Software Engineering

Adaptive Bandwidth Management and QoS Provisioning in Large Scale Ad Hoc Networks

Choosing a Content Delivery Method

A Novel Load Balancing Optimization Algorithm Based on Peer-to-Peer

5 Performance Management for Web Services. Rolf Stadler School of Electrical Engineering KTH Royal Institute of Technology.

How To Provide Qos Based Routing In The Internet

Disjoint Path Algorithm for Load Balancing in MPLS network

Bandwidth Management Framework for Multicasting in Wireless Mesh Networks

Implementation of Video Voice over IP in Local Area Network Campus Environment

Globule: a Platform for Self-Replicating Web Documents

Router Scheduling Configuration Based on the Maximization of Benefit and Carried Best Effort Traffic

AN APPROACH TOWARDS THE LOAD BALANCING STRATEGY FOR WEB SERVER CLUSTERS

Internet Protocol: IP packet headers. vendredi 18 octobre 13

AN OVERVIEW OF QUALITY OF SERVICE COMPUTER NETWORK

Assignment #3 Routing and Network Analysis. CIS3210 Computer Networks. University of Guelph

Building Secure Network Infrastructure For LANs

Analysis of Effect of Handoff on Audio Streaming in VOIP Networks

Content Distribution over IP: Developments and Challenges

Research on Errors of Utilized Bandwidth Measured by NetFlow

Infrastructure for active and passive measurements at 10Gbps and beyond

Internet Content Distribution

The QoS of the Edge Router based on DiffServ

Performance Evaluation of VoIP using Shortest-Widest and Modified Widest-Shortest QoS Routing Algorithms

IPv6 Fundamentals Ch t ap 1 er I : ntroducti ti t on I o P IPv6 Copyright Cisco Academy Yannis Xydas

Peer to Peer Search Engine and Collaboration Platform Based on JXTA Protocol

Request Routing, Load-Balancing and Fault- Tolerance Solution - MediaDNS

Research on P2P-SIP based VoIP system enhanced by UPnP technology

Performance Comparison of Server Load Distribution with FTP and HTTP

Monitoring Large Flows in Network

IPTV AND VOD NETWORK ARCHITECTURES. Diogo Miguel Mateus Farinha

Chapter 2. Literature Review

Giving life to today s media distribution services

Smart Queue Scheduling for QoS Spring 2001 Final Report

Implementation of a Lightweight Service Advertisement and Discovery Protocol for Mobile Ad hoc Networks

Transcription:

ISBN 978-952-5726-00-8 (Print), 978-952-5726-01-5 (CD-ROM) Proceedings of the 2009 International Symposium on Web Information Systems and Applications (WISA 09) Nanchang, P. R. China, May 22-24, 2009, pp. 444-448 A Random Selection Algorithm Implementing Load Balance for Anycast on Application-layer Zhiguo Zhou 1,Gaochao Xu 2, Chunyan Deng 2,Jinxin He 3 Jianhua Jiang 4 1 College of Computer Northeast Normal University Changchun,Chian 2 College of Computer Science and Technology Jilin University Changchun,China 3 College of Earth Sciences Jilin University Changchun,China 4 Department of Information Changchun Taxation College Changchun,China zhouzg281@nenu.edu.cn,{xugc,dengcy}@jlu.edu.cn,hejx@jlu.edu.cn,jianhuajiang@yahoo.com Abstract The key point of anycast on application-layer is how to select an optimal server for clients from a group of anycast servers with the same address. Sometime, several concurrent requests may be connect to the same server, which will lead to the dropping of the quality of service (QoS). And great network load fluctuation will occur, too. A random selection algorithm is proposed in the paper. In the algorithm, when concurrent requests occur, a server will be selected randomly for clients from the optional servers which accord with anycast condition. The method not only ensures the QoS of anycast server but also provides clients with an optimal server. The algorithm has been proved to be feasible and efficient by simulation experiment. Anycast Member Server A Anycast Packets Router Anycast Member Server B Anycast Packets Index Terms Anycast; Application-layer; Random Selection Algorithm; Load Balance Client 1 Client 2 I. INTRODUCTION With the rapid development of computer technology and communication technology, more and more people share and access information on Internet. All kinds of multimedia applications such as audio communication, video communication, online entertainment and network games have all brought about high volume data, which has led to slow speed of network and the dropping of the QoS (quality of server). In order to offer enough network service, enhance the availability of service and improve the flow distribution of network, it is necessary to research new protocols and models to satisfy with the growing requirement. Anycast is just proposed on such basis. Anycast is a new communication mode in the standards of IPV6, which allows clients access the best or nearest server from a group anycast server with the same address, according to some performance or policy criteria. The communication model is shown in Figure 1. Anycast was proposed firstly by Craig Partridge [1] in RFC1546 in 1993, in which anycast was just be described as an experimental service. In RFC1546, anycast was defined to try to send messages to one server with anycast address at least and send to only one host at the best situation. Anycast was not proposed as a protocol. Because more and more application required such service, anycast was defined as a standard service model formally in RFC2373 [2] in 1998, in which IPv6 has allocated addresses to anycast and some of them are reserved for specific application of future. There are so many advantages of anycast. For example, it could solve the problem of the network congestion caused by the unbalanced flow distribution efficiently, Figure 1. Anycast model improve the load topology of network, balance the load of network, widen the service area and improve the QoS [3]. So as soon as anycast was proposed, researchers home and abroad paid high attention on it. Now, most research focuses on anycast on application-layer and network-layer, the design and implementation of anycast model and the analysis and design of anycast routing algorithm. II. RELATED RESEARCH Anycast service of IPv6 includes the service on application-layer and on network-layer. The service on application-layer includes website mirror, server replication, DNS and Network Time Protocol (NTP), etc. The technology of replicating Web service is a common approach to improve the efficiency and extensibility of service and to satisfy with the QoS of a large number of clients. It not only plays an important role on parallel computing but also becomes the key technology of services such as website mirror, DNS and network games. According to reference [4] and [5], S. Bhattacharjee proposed an anycast communication paradigm of server replication on application-layer. They examine the definition and support of the anycasting paradigm at the application layer, providing a service that maps anycast domain names into one or more IP address using anycast resolvers. In addition to being independent from networklayer support, their definition includes the notion of filters, functions that are applied to groups of addresses to affect the selection process. They consider both metric-based filters (e.g., server response time) and policy-based filters; they further allow filtering both at the anycast resolver and local to the anycast client. A key input to the filtering 2009 ACADEMY PUBLISHER AP-PROC-CS-09CN001 444

process is metric information describing the relative performance of replicated servers. They examine the use of various techniques for maintaining this information at anycast resolvers. Chris Noble proposed an optimal algorithm, with which all anycast requests of clients could be mapped to the optimal video server [6]. For example, Video on Demand (VOD) service was just such kind of application. The algorithm was an application-layer anycast algorithm based on economical model and queuing theory. Clients could be satisfied to the greatest degree so as to get the highest use factor of system resource with this algorithm. Chia-jung Li proposed an efficient anycast scheme for discovering K Services in mobile ad-hoc networks [7]. In the Anycast-K scheme, an anycast tree based on the virtual backbone was established to reduce unnecessary message transmission. Every node could know the service status including hop count and service number, according to which the clients selected the optimal service but not selected blindly. Heng Chang studied a set of efficient algorithms for server selection under the condition of imprecise network delay using application-layer anycast [8]. The algorithms used ticket-based probing approach to search the path with short delay and probed the server load at the same time so as to select the optimal server with short delay and minimum load. Gaurav Agarwal proposed CDAA (Content Distribution Architecture using Anycast) [9] that used server replication and network-layer anycast to direct clients to the optimal server. CDAA need only minimal bandwidth and little computing requirements. It was scalable, extendable and transparent to current network applications and protocols. It assigned the same anycast address to content-equivalent servers on both applicationlayer and network-layer. CDAA adopted anycast as the communication mode between clients and replication servers, which improved the QoS and user satisfaction and avoided the efficiency losing caused by clients blind selection. The other application of anycast is load balancing. It is an important method to improve distributing computing because it could reduce network congestion, improve the use factor of resource and avoid single invalidation. Recent research mainly focus on how to implement load balancing through NAT on IP-layer, port listening on TCP-layer, DNS on application-layer and HTTP redirection [10]. The research on load balancing of uncertain host has achieved greatly. Yamamoto M proposed a network paradigm using active anycast to implement server load balancing [11, 12]. In the paradigm, active anycast was defined as a communication mode of active network [13]. In active anycast, client only sent requests for a server with anycast address. Then active router would select an adequate server according to load balancing, change the anycast address to unicast address then transmit. In addition, Miura H proposed a server selection policy using RTT information measured at an active router [14]. By means of the server selection based on this RTT, the selection of the server considered not only the server load but also network congestion and routing path length. III. LOAD FLUCTUATION The most important reason of network congestion is that the load is greater than the capacity of network, which leads to longer packet delay, increasing losing ratio and the decline of application capability. The most important effect of application-layer anycast lies on decreasing network congestion, implementing load balancing, improving the efficiency of the use of resource and avoiding single invalidation so as to improve the QoS of network. When implementing application-layer anycast, client only needs to send requests for a server with anycast address. Then center server or router will select an optimal server according to the standard of load balancing, change the address of destination to corresponding unicast address then transmit. The selection standard of center server is decided by the load, the information of RTT, hop count, routing path length and other parameters of every anycast server. It is changes in different algorithm. In application-layer anycast, anycast parser always leads clients to connect to the optimal server, which looks perfect. However, a potential problem of load fluctuation [15] will occur. According to the routing length, packet being sent to anycast address will be routed to the nearest server, which may be not the optimal server. From above we know that there are obvious disadvantages in current protocol. Once network congestion occurs in the nearest server, it can t be detected and solved in time in current anycast routing protocol, which will make the congestion even more severe and make the capacity of network decline. Anycast parser may lead more than one client to connect to a server when it detects an adequate server. It will make the server load be heavy immediately and can t respond to all other clients. In the mean time the capability of the server declines, other servers are idle. When anycast parser refreshes parameters of all servers, it will detect the other adequate server and lead many clients to connect to that server. For the same sake, the capability of the latter server will decline immediately, too. And the problem of heavy load of former server will be solved. As is called load fluctuation, which is shown in Figure.2. 445

P1 M Pn A R1 G Figure 2. The communicating mode of application-layer anycast In Figure.2, M is a central server, and A~G are anycast servers, and R1, R2 and R3 are routers, and P1~Pn are clients. It is supposed that the load of B~G is similar and there are ten clients connecting to each of them at one time. Only one client connects to A. At this moment, n clients all request to M. According to the load algorithm, M will detect that A is the adequate server, so M will transmit all requests to A. Thus load fluctuation occurs. The load of A becomes heavy suddenly, and the QoS declines immediately, too. The worst situation is server down. Such load fluctuation affects the VOD, network game sever greatly. Because of load fluctuation, the communication between server and client will be influenced. For server, the load will become unbalanced. A server may be very busy at one moment and be idle at another moment, which doesn t benefit the utilizing of resource in probability. For client, the establishing time of communication will be prolonged because so many clients try to connect to the same adequate server at the moment, which is fatal to the real-time server. In a word, load fluctuation not only reduces the use factor of network resource but also make it difficult to ensure the quality of anycast communication. IV. RANDOM SELECTION ALGORITHM The solution to load fluctuation of anycast communication is to balance load of all servers. In other words, anycast parser selects several optional servers with similar load and transmits requests from different clients to these servers. Through this method, each client could get adequate resource in time. It could also avoid load fluctuation caused by frequent change of demand-andsupply on resource point. A random selection algorithm is proposed in this paper. It is introduced into the server selection policy of anycast communication on application-layer. In this algorithm, anycast prototype is modified partly. The most important modification is that anycast parser doesn t select adequate server directly for client any more. According to the requests of clients, anycast parser detects several optional servers with low load and selects one of them randomly to the client. Then it transmits the connection to the anycast server. In Figure.2, it is supposed that there are n anycast servers in the anycast group. At the moment, the load value of each server is L1,L2,,Ln. So, the average load value of them could be described as L= B R2 F C R3 D E (L1+L2+ +Ln)/n. When P requests to M, M will count the load value of each anycast server respectively. There are many algorithms of calculating network load [16]. The indicator includes server response time, bandwidth of network (unidirectional bandwidth or bidirectional bandwidth),connection number of clients, CPU load of server, the utilization of disk, the utilization of memory and process number, etc. According to the values of Li and L, Li is divided to two sets: P{P1, P2,,Pm} and Q{Q1,Q2,,Qk}(1 m, k n, m+k=n. P is a server set whose load is below the average load. Q is the one over the average load.). According to random selection algorithm, the central server will select a server from the set of P randomly and respond the address of this server to client. For client, the responded server may not be the optimal one. However for anycast group and its members, it could reduce load fluctuation caused by concurrency requests from several clients greatly. A. Algorithm description According above discussion, the core algorithm could be described as following: Calculating the value of load connecting the client to all servers and storing it to array L Counting the sum of array L and calculating its average value Dividing array L to two sub arrays, which is P and Q. P is a server set whose load is below the average load and Q is the one over the average load. Selecting a value in P randomly and choosing the corresponding server as the adequate server to client Responding the unicast address of P [Random_Num] to client Client requesting to server with this unicast address and starting communication after the connection is established. B. Experiments The author experiments on above algorithm. Figure 3 shows load figure of M without using random selection Figure 3. Load figure without using random selection algorithm Figure 4. Load figure using random selection algorithm 446

algorithm. In this figure, time is selected as horizontal axis and load (connection number of TCP) is selected as vertical axis. Figure 4 shows load figure of the same server using the algorithm. Contrasting to Figure.3, the effect of the random selection algorithm in Figure.4 is perfect. In Figure.3, load fluctuation is larger and the connection number always increases suddenly. However, such fluctuation tends to smooth in Figure.4. C. Algorithm improvement Through experiments, the author finds that in general the number of member whose load is below the average load amounts to almost half of the whole. It means that the selection scope is so wide that the selected anycast server may be far away from the actual optimal server. If the selection scope could be reduced, load fluctuation could be avoided, and the more adequate server may be selected, too. So, the author proposes the following two improving methods: (1) Selecting 0.75L as the section point when dividing L to P and Q. Through this method, the length of array P becomes shorter, and the one of Q becomes longer accordingly. The reduction of member number of P means the reduction of the selection scope and the increasing rate of selecting optimal server. So, load fluctuation is avoided and the selected sever is more close to the optimal server. The experiment data is shown in Figure.5. (2) Selecting a designated value which could be input from keyboard as the section point when dividing L to P and Q. For example, the author selects 0.85L in the experiment. The experimental result is shown in Figure.6. Contrast to Figure.4, the effect of Figure.5 and Figure6 is better. But there are not obvious difference between Figure.5 and Figure.6. From analysis above, we know that the selection of section point is crucial. It would affect the grade of load balance of server and the quality of anycast communication. (1) If a large value is selected as section point, the length of P and Q will be almost equal. For example, if we select L, the average load value, as section point, it means that almost half of servers are optional. When parser selects one of them as the responding server, the scope is so wide that the server may be far away from the optimal server though load fluctuation might be reduced. The longest length is L-L_MIN in theory. It would affect the QoS of anycast and go against the grain of anycast because client couldn t get a better server, to say nothing of the best one. (2) If a small value is selected as section point, the length of P will be shorter than the one of Q. In other words, the selection scope will decrease and the rate of get the optimal server will increase. A very small section point may lead to only one optional server in P so the Figure 5. Load figure of selecting 0.75L as section point server is just the optimal server. However, load fluctuation discussed above would occur. The connection number increases suddenly and the QoS decreases immediately, which may even lead to server down. Figure 6. Load figure of selecting a designated value (0.85L) as section point V. CONCLUSIONS Many factors such as the use factor of CPU, memory and network and the connecting number of TCP all affect the capacity of server. The measurement is decided by the service type. For example, servers of NTP, network game and WWW mirror access little CPU and memory, but they are greatly influenced by bandwidth, the use factor of network and the number of TCP connection. By contraries, CPU and memory affect the performance of computing server greatly. So the selection of section point of load is crucial to the algorithm proposed in this thesis. And how to compute the load will influence the effect of the algorithm, too. ACKNOWLEDGMENT This research is supported by Science Foundation for Young Teachers of Northeast Normal University (No. 20081004) *Corresponding author : Gaochao XU(xugc@jlu.edu.cn) REFERENCES [1] Partridge C, Mendez T, Milliken W. Host Anycasting Service. RFC1546. (Nov.1993) http://www.ietf.org/rfc/rfc1546.txt [2] HINDEN R, DEERING S. IP Version 6 Addressing Architecture.RFC2373. (July 1998). http://rfc.net/rfc2373.html [3] P. Danzig, D. Delucia, and K. Obraczka. Massively Replicating Services in Wide-Area Internetworks. Univ. Southern California, Los Angeles, CA, Tech. Rep. pp.93-541, 1994. [4] Ellen Zegura, Mostafa Ammar, Zongming Fei, Samrat Bhattacharjee. Application-layer Anycasting:A Server selection Architecture and Use in a Replicated Web Service. IEEE/ACM Transactions on Networking, Vol.8, No.4, pp. 455-466, 2000 [5] S Bhattacharjee, MH Ammar, EW Zegura, V Shah. Application-layer Anycasting. Proc. IEEE INFOCOM'97, 1997 [6] Z. D. Wu, C. Noble, D. Huang. Optimal Video Distribution Using Anycasting Service. Proc. of INET99, June 1999 [7] Chia-jung Li. An Efficient Anycast Scheme for Discovering K Services in Mobile Ad-hoc Networks. TaiBei:Feng Chia University.2005 [8] Heng Chang, Weijia Jia, Ling Zhang. Distributed server selection with imprecise state for replicated server group. 7th International Symposium on Parallel Architectures, Algorithms and Networks.Los Alamitos:IEEE Computer Society, pp. 73-78,2004 447

[9] Gaurav Agarwal, Rahul Shah, Jean Walrand. Content distribution architecture using network layer anycast.los Alamitos: IEEE Computer Society, pp.124-132, 2001 [10] Xue Jun,Li Zeng-zhi,Wang Yun-lan. Development of Technology of Load Balancing.Mini-Micro Systems, Vol.24, No.12, pp.2100-2103, 2003 [11] Yamamoto M,Miura H,Nishimura K, H Ikeda. A Network-Supported Server Load Balancing Method: Active Anycast.IEICE Transactions on Communications, Vol.E84-B, No.6, pp.1561-1568, 2001 [12] Miura H,Yamamoto M,Nishimura K,et al. Server load balancing with network support:active anycast.proceedings of International Working Conference on Active Networks. Berlin: Springer-Verlag, pp.371-384, 2000 [13] Kalamullah Ramli. An improved active network concept and architecture for distributed and dynamic streaming multimedia environments with heterogeneous bandwidths. Germany, University Duisburg-Essen, 2003. [14] Miura H, Yamamoto M. Server selection policy in active anycast.ieice Transactions on Communications, Vol. E84-B,No.10, pp.1-4,2001 [15] Michele Colajanni, Philip S. Yu. Adaptive TTL schemes for load balancing of distributed Web servers. ACM SIGMETRICS Performance Evaluation Review, Vol.25, No.2, pp.36-42, 1997 [16] Deng X.; Liu H.N.; Long J.; Xiao B.. Competitive Analysis of Network Load Balancing. Journal of Parallel and Distributed Computing, February 1997, Vol. 40, No.2, pp.162-172(11), 1997 448