|
|
|
Load Balancing And Yahoo
high-volume site like that the operating system
Yahoo! knows that the parameters and web server
actual quality of service software can be adjusted to
any web server provides to end achieve better performance.
users basically depends on
network-transfer speed and server An alternative approach is to
response time. Network-transfer improve performance by increasing
speed refers to the Internet-link the number of web servers. This
bandwidth while server-response approach would attempt to
time depends upon resources distribute traffic unto a cluster
including fast CPU, lots of RAM of back-end web servers that need
and good I/O performance. Once not be large-scale machines. Web
these resources are exhausted and server scalability is achieved
the web-server is encountering when more servers are added to
heavy traffic, a problem would distribute the load among the
surely arise. group of servers or server
cluster.
Load Balancing
This is what load balancing is
A problematic situation all about. It involves the fine
pertaining to difficulty in tuning of a computer system,
handling high volumes of incoming network or disk subsystem in
traffic can be solved either order to more evenly distribute
through installing more RAM on the data and/or processing across
existing machines or replacing available resources. Load
the CPU with a faster one. The balancing is distributing,
use of faster or dedicated SCSI processing and communications
controllers and disks with activity evenly across a computer
shorter access time can also be network so that no single device
done. Software can be tuned so is overwhelmed. Busy websites
usually use two or more web traffic and connections to a
servers in a load balancing website.
scheme so that when one server
gets overwhelmed with requests, Yahoo! was granted a patent from
traffic is forwarded to another a filing done in 1999 regarding
server with more capacity. coordinating information between
multiple servers that share
There are two probable reasons information as well as servers
why a company could want to load that may cache some of the
balance traffic across firewalls. information. Load balancing
One is for purely technical devices are becoming very common
reasons and the other is centered in supporting high-traffic
on winning business. The websites. These devices evolve as
technical aspect should be websites grow in terms of size,
quickly addressed as soon as complexity and traffic flow.
funds and environment allow.
The presence of multiple web
When there is only one web server servers in a server group
responding to all incoming HTTP requires that HTTP traffic be
requests for a website, it may evenly distributed among the
not be able to perform servers. These servers should
accordingly especially if the appear as a single web server to
website has gained popularity. the web client. The load balancer
Loading of web pages will be very simply intercepts each request
slow and some users would have to and redirects it to an available
wait for their requests to be server in the server cluster.
processed. It can come to a point
where upgrading the server Methods of Load Balancing
hardware is no longer cost
effective due to the increased Load balancing can be achieved in
a number of ways. Choice would servers which would result to
depend on the individual request not being sent to the
requirement, available features, load balancing DNS server.
complexity of implementation and
the cost. The user company would In Hardware Load Balancing,
have to determine its hardware load balances route
circumstances to determine which TCP/IP packets to various servers
option would work best. in a cluster. This method is said
to provide a powerful topology
The Round Robin DNS Load with high availability. It uses
Balancing is one of the early circuit level network gateway to
adapted load balancing route traffic. Its one downside
techniques. The built-in round is the higher cost incurred as
robin feature of BIND of a DNS compared to other methods.
server facilitates cycling
through the IP addresses The most commonly used method is
corresponding to a group of Software Load Balancing. Load
servers in a cluster. It is a balancers often come as an
fairly simple and inexpensive integrated component of expensive
method which is very easy to web server and application server
implement. However, its downside software packages. This method is
is that the DNS server does not more configurable based on
have any knowledge of server requirements and can incorporate
availability thus may continually intelligent routing base on
point to an unavailable server. multiple input parameters. An
It has the ability to additional hardware needs to be
differentiate by IP address but provided to isolate the load
not by server port. There is also balancers.
the possibility that the IP
address is cached by other name Algorithm of Server Load
Balancing that all servers are equally
assigned. The allocation is much
When HTTP requests are assigned orderly than random but it may
to any server picked randomly not be enough based on processing
among the group of servers, this overhead required and when there
is called random allocation. It are differences in server
is possible that one server may specification in a server group.
be assigned more requests than
the others, but generally each The shortcoming of the
server gets its share of the round-robin allocation has been
load. It can be very easy to eliminated by the weighted
implement but the risk of round-robin version. In this
overloading one while case, a server that is capable of
under-utilizing another is big. handling twice as much load as
the other can get a weight of
The IP sprayer assigns the two. This means that the IP
requests to a list of the servers sprayer will assign two requests
on a rotating basis when the to the powerful server as against
round-robin allocation is used. one request assigned to the
The first request goes to a weaker one. This takes care of
randomly picked server in a group the capacity of the servers in
so that the entire first request the group. However, it does not
need not go to the same server consider the advanced load
especially if more than one IP balancing requirements like
sprayer is involved. The circular processing time for individual
order is followed in redirecting request. An efficient load
the traffic for subsequent balancer should be capable of
requests. The server which has intelligent monitoring that would
been assigned a request moves to help it direct requests to the
the end of the list to ensure server that is more capable of
handling them.
About the Author:
http://www.theinternetone.net
Read more articles by: Danny Wirken
Article Source: www.iSnare.com
|
|
