High-availability web server clusters use redundancy to ensure continuous service if any single component fails. They distribute workload across multiple servers for improved efficiency and scalability. The document proposes using Linux Virtual Server to implement such a cluster, with virtual servers distributing requests to real servers via network address translation, direct routing, or IP tunneling. Hardware requirements include servers, switches, and cabling to physically set up the redundant infrastructure.

1. High-Availability
Web Server Cluster
Group 10

2. Group Member
Group Leader :
Wong Sing
Group Member :
CHAN Chun Cheong
CHAN Tsun Tat
NG Chun Lam
PUN Wai Yin

3. What is High-Availability cluster?
High-Availability clusters (Fail-over Clusters) are
implemented primarily for the purpose of improving the
availability of services that the cluster provides. The
operate by having redundant nodes, which are then used to
provide service when system components fails. High-
Availability cluster implementations attempt to use
redundancy of cluster components to eliminate single points
of failure.

4. Why High-Availability Web Server
Cluster is important? (1/2)
High availability
Redundancy server provide continued service when
single points of failure
Server loading
Via "Load Balancer" to distribute job to different
server to avoid single server overload, provide some
degree of improve efficiency and capacity

5. Why High-Availability Web Server
Cluster is important? (2/2)
Scalability
User is not feel change when increase or decrease
the redundant server in the cluster
All along the grow of Internet, increase redundant
server to improve the performance of cluster to deal
with added user. Rather than adopt single high-end
server

6. Potential Solution (1/2)
The Server side approach
IP-level scheduling approach
Linux Virtual Server <== we do it
Berkeley's MagicRouter, Cisco's LocalDIrector
Application-level scheduling approach
EDDIE, Reverse-proxy, pWEB, SWEB
Round-Robin DNS approach
DNS Server binding a host with multiple IP address
DNS Server response a rotation IP address
DNS Cache would ensure a specific web server deal with
the client

7. Potential Solution (2/2)
The client-side approach
Berkeley's Smart Client
A applet running at the client side which choose a
server based on the load information of all the
servers

8. Implement High-Availability Server
Cluster using Linux Virtual Server
Architecture of a generic virtual server

9. What is Linux Virtual Server?
Linux Virtual Server System is a group of a virtual server and
few real server to a number of real server
Number of real server decide on how to implement
More real server indicate higher capability to deal with user
Virtual Server is a highly scalable and highly available server
built on a cluster of real servers. Virtual server is a entry-pint
of server cluster system. Also, various alias (Load Balancer,
Linux Director) of virtual server indicate how to dispatch
requests to different real server
Virtual Server appear as a host externally

10. High Availability System
HA distributed 2 mode
1. real server
2. virtual server

11. Linux Virtual Server Architectures
Virtual Server via NAT
Virtual Server via IP Tunneling
Virtual Server via Direct Routing

12. Virtual Server via NAT
Virtual Server (Load Balancer) response NAT as router
Record the socket of a new web request and assign to a
real server via specific scheduling
Then, Virtual Server map a follow request of it to same
real server
Virtual server focus on NAT in many time. The Virtual
server may be a new bottle-neck of the whole system.
Therefore, Virtual Server handle around 20 real server
at most
As a result of Virtual Server change the detestation of
IP address and the port of request, the listen port of
Real Server can differ

13. Virtual Server via Direct Routing
Virtual Server (Linux Director) and Real Server share a
same Virtual IP address
Virtual Server re-write the destination MAC address of
frame re-direct to Real Server only
As a result of same Virtual P address of Virtual Server
and Real Server. Thence, it must disable ARP response
of Real Server to prevent Real Server harass Virtual
Server work properly
ARP issues occur on Virtual Server via IP Tunneling too
The listen port of all Real Server must be same and
Virtual via IP Tunneling too

14. Virtual Server via IP Tunneling
Virtual Server (Load Balancer) encapsulate a new request
(packet) within IP datagrams and redirect to a Real
Server
Then, a real server replies to web client directly
Virtual Server is do not need to NAT. Thence, Virtual
Server handle more Real Server than Virtual Server via
NAT
In addition, Real Server can be geographically
distributed in any network. Rather than limited in LAN

15. Failure on Load Balancer (Virtual
Server)
The Load Balancer might become a single failure point of
the whole system
Redundancy load balancer can assuage this issues
Backup Load Balancer will take job from primary Load
Balancer when failure of the Primary Load Balancer
Via heartbeat daemon run on both, they inform other
"I'm alive" periodically
If Backup Load Balancer cannot receive heartbeat in the
specific time, Backup Load Balancer will be Primary Load
Balancer

16. Limit of Linux Virtual Server
As a result of Layer 4 Load Balancing, it is regard
different information to distribute request to Real
Server compare with Layer 7 Load Balancing.
When Layer 7 Load Balancing compare with Layer 4 Load
Balancing, it is more flexible than and more complex than

17. Logical Topology
Network Client 192.168.1.0/24
Network DNS 10.0.0.0/8
Network Linux Virtual Server 172.16.0.0/12

18. Requested Hardware (1/3)
Real Server & Virtual Server
Lenovo ThinkCentre A63 Tower
Form Factor Tower
Processor AMD Athlon x 2 245, 2.90GHz, 2MB L2Cache
Memory 2GB, DDR 3 1066 MHz
HDD 320 GB 7200 rpm
NIC Marvell Gigabit Ethernet
Device DVD-ROM
PCI Expree x1 2 Full Height
USB 8
Warranty One year parts and labour
Price $ 3,080
Quantity 6 = 1 Client + 2 VS + 2 RS + 1 Router/DNS

19. Requested Hardware (2/3)
DNS Server and Router
Hardware configuration approximate Real Server &
Virtual Server
Need 2 connecter additionally
Realtek RTL8111B Gigabit Ethernet Controller
PCI Express 1.0a bus interface
Crossover Detection & Auto-Correction
Price $120 (estimate)
Quantity 2

20. Requested Hardware (3/3)
Network devices
Planet GSD-803 Switch
Port Density 8
Wire speed Gigabits
MAC Table Store 4K
Price $ 350
Intelligent scales power based on cable length
Cable
Cat 5e twisted pair
100 M $300
RJ 45 connector
20 $10

21. Project Budget
Item Quantity Unit Price
Lenovo ThinkCentre A63 6 $ 3,080
Realtek RTL8111B Gigabit NIC 2 $ 1,50
Planet GSD-803 Switch 1 $ 3,50
Cat 5e twisted pair 1 $ 3,00
RJ 45 connector 1 $ 10
Total $ 19,440

22. Milestone

23. Summary
We using Linux Virtual Server to implement High-
Availability Web Server Cluster. To provide a high-
availability, reliable, scalability, load-balancing Solution for
improve efficiency, productivity.
Business can upgrade their web server without revoke
original web server. In the upgrade process, without need to
suspend the web service and the user would not feel the
effect.
Finally, HA Web Server Cluster contributes a bit
of capability to make Internet more tough