Transcript DYNAMIC LOAD BALANCING ON WEB

DYNAMIC LOAD BALANCING
ON WEB-SERVER SYSTEMS
by
Valeria Cardellini
Michele Colajanni
Philip S. Yu
Contents







INTRODUCTION
CLIENT-BASED APPROACH
DNS-BASED APPROACH
DISPATCHER-BASED APPROACH
SERVER-BASED APPROACH
COMPARING THE APPROACH
CONCLUSIONS
INTRODUCTION


disproportionate increase in client requests to
popular Web sites
solution

mirroring



replicate information across a mirrored-server system
not user-transparent
distributed Web-server system

Information can be distributed among server nodes:



Successful load-balancing



1st : content tree replication – LAN, WAN
2nd : information sharing – LAN
transparent users
appear as a single host to outside world
four distributed Web-server architectures


Client-based
DNS-based , dispatcher-based , server-based
CLIENT-BASED APPROCH

Web Clients

Web client selects a node of the cluster and submits the
request to the selected node



Smart Clients



Migrates server functionality to the client through a Java applet
Increase network traffic and network delay
Client-Side Proxies


Netscape Navigator – random select
Limited practical applicability and is not scalable
Web Cluster standpoint, proxy servers are similar to clients
not universally applicable
DNS-BASED APPROACH



The cluster DNS translates URL to the IP address
User transparent
ex) www.yahoo.com www.cnn.com
DNS-BASED APPROACH (con’t)

Drawbacks

The DNS a limited control on the request reaching the Web
cluster


scheduling algorithm that the cluster DNS uses to
balance the Web-server node’s load

constant TTL algorims





Between the client and the web server DNS, many
intermediate name servers can cache the logical name to IP
address mapping to reduce network traffic and every web
browser typically caches some address resolution
System-stateless
Server-state-based
Client-state-based
adaptive TTL algorims
DNS-Based Architecture Comparison
DISPATCHER-BASED APPROACH


Centralize request scheduling and completely control
client-request routing
Request routing among server is transparent-unlike
DNS-based




DNS deals address at the URL level, the dispatcher has a single,
virtual IP address(IP-SVA)
Dispatcher uniquely identifies each server in the
system through a private address
Dispatcher typically use simple algorithms to select
the Web server
difference by routing mechanism




Packet Single-Rewriting
Packet Double-Rewriting
Pachet Forwarding (network dispatcher, One-ip address)
HTTP redirection
Packet Single-Rewriting


TCP router acts as an IP address dispatcher
High System availability
Packet Double-Rewriting

Two solution using this approach
 Magicrouter

round-robin, random, incremental load algrithm
 Cisco System’s Local Director

least number of active connections
Packet Forwarding

Network Dispatcher



Dispatcher forward packets to the selected server
using its physical address without IP modification
Level 1 : single-rewriting mechanisim
Level 2 : LAN network dispatcher
Packet Forwarding (con’t)

ONE-IP address
 uses the ifconfig alias option to configure a
Web-server system with multiple machines
 implemented with two techniques


Routing-based dispatching
Broadcast-based dispatching
HTTP Redirection



among the Web-server nodes through the
HTTP’s redirection mechanism
no IP address modification
two techniques
 Server-state-based dispatching

used by Distributed Server Groups architecture
 Location-based dispatching


used by Cisco System’s DistributedDirector appliance
Dispatcher-Based Architecture Comparison
SERVER-BASED APPROACH

Use two level dispatching mechanism



Integrating the DNS based approach with redirection
techniques executed by Web server
Solves most DNS scheduling problem
Two Solution

HTTP redirection

Packet redirection
COMPARING THE APPROACHES

Approach Trade-off Summary
COMPARING THE APPROACHES

Performance Evaluation
CONCLUSIONS


network bandwidth can constrain loadbalancing performance
LAN-distributed Web-server cluster are thus
a limited solution to increased client requests