Design and implementation of an MPLS based load balancing architecture for Web switching

Transcription

1 School of Electronic Engineering - DCU Switching and Systems Laboratory Design and implementation of an MPLS based load balancing architecture for Web switching Radu Dragos, Sanda Dragos and Martin Collier dragosr@eeng.dcu.ie dragoss@eeng.dcu.ie collierm@eeng.dcu.ie Page 1 of 17

2 1. WWW - The preferred technology: web server limitations - New architectures - Web clusters & New technologies - Web Switching MultiProtocol Label Switching: The preferred technology for: - Traffic engineering - Switching The next generation of networks Design & : an MPLS web switching architecture for QoS enabled networks Page 2 of 17

3 1. WWW - The preferred technology: web server limitations - New architectures - Web clusters & New technologies - Web Switching MultiProtocol Label Switching: The preferred technology for: - Traffic engineering - Switching The next generation of networks Design & : an MPLS web switching architecture for QoS enabled networks Page 2 of 17

4 1. WWW - The preferred technology: web server limitations - New architectures - Web clusters & New technologies - Web Switching MultiProtocol Label Switching: The preferred technology for: - Traffic engineering - Switching The next generation of networks Design & : an MPLS web switching architecture for QoS enabled networks Page 2 of 17

5 1. WWW - The preferred technology: web server limitations - New architectures - Web clusters & New technologies - Web Switching MultiProtocol Label Switching: The preferred technology for: - Traffic engineering - Switching The next generation of networks Design & : an MPLS web switching architecture for QoS enabled networks Page 2 of 17

6 2. Client 3 Client 1 Client 2 Requests Web Server Html CGI scripts File server Application server Multimedia Streaming Database server E-commerce server Client 4 Packet Dropped packet Page 3 of 17

7 3. Solutions to the problem of over-congested Web servers 3.1. Web Content Caching Requests Client 1 Client 2 Web Cache Cached Web Content Web Server Client 3 Client 4 Page 4 of 17 Limitations: static vs dynamic content

8 3.2. Web Content Mirroring Mirror 2 Client Mirror 1 Main Web Server Mirror 3 Drawbacks: user decision Mirror 4 Page 5 of 17

9 3.3. Cluster of web servers Client 1 Client 2 Dispatcher Web server 1 Web server 2 Web server 3 Client 3 Web server 4 Categories: - replicated content - mirrored content Page 6 of 17

10 Overloading a web server λ 10connections/second 1/µ 30seconds 1 server: C servers: C servers: C Page 7 of 17

11 Overloading a web server λ 10connections/second 1/µ 30seconds connections X Graph server1 secconds 1 server: C servers: C servers: C Page 7 of 17

12 Overloading a web server λ 10connections/second 1/µ 30seconds connections X Graph server1 secconds 1 server: C connections X Graph server1 server2 2 servers: C secconds 3 servers: C Page 7 of 17

13 Overloading a web server λ 10connections/second 1/µ 30seconds connections X Graph server1 secconds 1 server: C connections X Graph server1 server2 2 servers: C secconds connections X Graph server1 server2 server3 3 servers: C secconds Page 7 of 17

14 4.2. The TCP continuity problem TCP Connection 1 Dispatcher Web server 1 Web server 2 Client TCP Connection 2 Web server 3 Web server 4 Page 8 of 17

15 Round-Robin DNS DNS req: web server 1 IP 1 Client 1 DNS reply: IP 1 Client 2 DNS req: DNS reply: IP 2 DNS web server 1 IP 2 DNS req: Client 3 DNS reply: IP 3 web server 1 IP 3 Limitations: - DNS caching - IP browsing - load information Page 9 of 17

16 5.2. HTTP redirect Cluster request Client request redirection Main Server Server Server Drawback: - redirection delay Page 10 of 17

17 5.3. Load balancing switches Web switch Web server 1 Web server 2 Web server 3 Backup Web switch Drawbacks: - single point of failure - expensive solution Web server 4 Page 11 of 17

18 to MPLS Ingress node LER MPLS cloud LER Egress node LER LER MPLS node = LSR LSP MPLS edge node = LER LSR Label Switching Hop LSR LER LSR - Label Switch Router LER - Label Edge Router LSP - Label Switching Path Page 12 of 17

19 6.2. Traffic Engineering objectives for MPLS environment - traffic oriented - resource oriented Page 13 of 17

20 6.2. Traffic Engineering objectives for MPLS environment - traffic oriented - resource oriented Page 13 of 17

21 6.2. Traffic Engineering objectives for MPLS environment - traffic oriented - resource oriented Relation with load balancing: Page 13 of 17

22 6.2. Traffic Engineering objectives for MPLS environment - traffic oriented - resource oriented Relation with load balancing: resource oriented objective - server over- or under-utilization Page 13 of 17

23 6.2. Traffic Engineering objectives for MPLS environment - traffic oriented - resource oriented Relation with load balancing: resource oriented objective - server over- or under-utilization traffic oriented objective - supporting QoS operation Page 13 of 17

24 7. IP pkt LER IP pkt L syn L a Servers Client IP pkt L 1 IP pkt L syn L a LSR LSR DISPATCHER Client MPLS cloud Client LER Page 14 of 17

25 7. IP pkt LER IP pkt L syn L a Servers Client IP pkt L 1 L b IP pkt L syn L a IP pkt L 1 IP pkt L 1 LSR IP pkt L 1 L b LSR DISPATCHER Client MPLS cloud Client LER Page 14 of 17

26 8. LER MPLS cloud WebServer 1 DISPATCHER WebServer 2 LER LER Linux Box WebServer n - IP Client (not MPLS capable) DISPATCHER - Host UML MLPS capable Linux router - Linux Firewall WebServer - UML Edge MPLS capable Linux server LER - Edge MPLS capable Linux router Page 15 of 17

27 9. Performance evaluation Page 16 of 17

28 9. Performance evaluation Round-robin load balancing for large files: Server 1 Server 2 Server3 no. % no. % no. % 20 con 7 35% 5 25% 8 40% 30 con % 12 40% % 50 con 16 32% 16 32% 18 36% Page 16 of 17

29 9. Performance evaluation Round-robin load balancing for large files: Server 1 Server 2 Server3 no. % no. % no. % 20 con 7 35% 5 25% 8 40% 30 con % 12 40% % 50 con 16 32% 16 32% 18 36% Round-robin load balancing for small files: from server 1 from server time (s) Server 1 Server 2 Server3 no. % no. % no. % 300 con % 99 33% 99 33% λ = con % % % λ = con % 96 32% % λ = 12 Table 2: Round-robin load balancing for small files Page 16 of 17 buted requests 6 and future work

30 10. exploiting the QoS capabilities of underlaying network a novel technique for the next generation switching protocols (e.g. MPLS) a working, cost effective architecture in an open source environment MPLS based solution performed well even for highly loaded Web switches (12 connections/second 1 million hits/day) porting to hardware can increase the performance Page 17 of 17

31 10. exploiting the QoS capabilities of underlaying network a novel technique for the next generation switching protocols (e.g. MPLS) a working, cost effective architecture in an open source environment MPLS based solution performed well even for highly loaded Web switches (12 connections/second 1 million hits/day) porting to hardware can increase the performance Page 17 of 17

32 10. exploiting the QoS capabilities of underlaying network a novel technique for the next generation switching protocols (e.g. MPLS) a working, cost effective architecture in an open source environment MPLS based solution performed well even for highly loaded Web switches (12 connections/second 1 million hits/day) porting to hardware can increase the performance Page 17 of 17

33 10. exploiting the QoS capabilities of underlaying network a novel technique for the next generation switching protocols (e.g. MPLS) a working, cost effective architecture in an open source environment MPLS based solution performed well even for highly loaded Web switches (12 connections/second 1 million hits/day) porting to hardware can increase the performance Page 17 of 17

34 10. exploiting the QoS capabilities of underlaying network a novel technique for the next generation switching protocols (e.g. MPLS) a working, cost effective architecture in an open source environment MPLS based solution performed well even for highly loaded Web switches (12 connections/second 1 million hits/day) porting to hardware can increase the performance Page 17 of 17

35 10. exploiting the QoS capabilities of underlaying network a novel technique for the next generation switching protocols (e.g. MPLS) a working, cost effective architecture in an open source environment MPLS based solution performed well even for highly loaded Web switches (12 connections/second 1 million hits/day) porting to hardware can increase the performance Page 17 of 17