Evaluating Network
Performance
THE PERFORMANCE OF ANY SINGLE
NETWORK
COMPONENT (OR NETWORK NODES) CAN BE
CHARACTERIZED BY MEANS OF THE
FOLLOWING
ATTRIBUTES:
PERFOMANCE ATTRIBUTES
i. Workload: The amount of work per unit of time
encountered by the component. Workload may be
characterized as requests per second or in terms of inbound
bandwidth such as packets per second.
ii. Response: The amount of time it takes for the component
to respond to a request.
iii. Loss Rate: The fraction of requests that do not receive a
response or receive an erroneous response.
iv. Throughput: The amount of work per unit of time that
receives a normal response through the component.
Sometimes it is also referred to as goodput.
v. Capacity: The maximum possible throughput of the
component.
vi. Utilization: The fraction of time the component is busy
General Principles for Performance Evaluation
Utilization Law
The utilization law states that the average utilization of any component in the computer
system is the throughput of the system multiplied by the amount of
service time each request requires at the server.
The utilization law is a simple statement of relationship between two performance
metrics that will be valid for any computer system when measured over a
reasonably large time period.
The throughput of the system when measured over this time period will be the number
of completed tasks divided by the time period. The time during this period that the
component is busy can be obtained by multiplying the number of completed tasks by
the average service time per task.
Utilization Law……
The utilization is the ratio of the busy period of the
component over the time period and can be readily seen to
satisfy the utilization law.
The utilization law can be used for determining what size or
capacity of a component to acquire for any computer system.
Suppose we want to keep a server at a web site to have a
utilization of no more than 50%, and we know that it takes an
average of 50 ms to process each request at the server.
In that case, the system throughput will be 10 requests/s.
As another example, let us consider a router which is
processing packets at a rate of 100 packets per second. Each
packet processing takes 2 ms. The utilization of the router is
going to be 20%.
Little’s law
Little’s law states that the average number of
pending requests in any computer system equals the
average rate of arrival of requests multiplied by the
average time spent by the request in the system.
Little’s law Cont..
In other words, if we take the average latency of a computer system and
multiply it with the average throughput of the system, we will get the
average number of requests in the system.
Little’s law does require the system to be stable (i.e., the actual
throughput needs to be less than the capacity of the computer
system) and requires averages over a reasonably large period.
Little’s law is valid for any type of queuing system where
people are serviced first come first served, not just for
computer systems.
There are some obvious uses of the law in planning for
computer systems.
Little’s law Cont..
The following examples illustrate some of the possible
uses of Little’s law in making planning decisions
regarding computer systems.
Example 1: Gaudy Gadgets Limited is rolling out its
web site and is expecting to draw a hit rate of 100
requests/s. Gaudy Gadgets wants to maintain an
average of 250 ms in the system for each request.
The web-hosting software is designed so that each
thread in the software handles a web request, and a
server is capable of running up to 10 threads with
satisfactory performance.
Little’s law Cont..
How many servers should Gaudy Gadgets used for its
web site?
Answer:
The average number of active requests we would
expect in the
system would be 250 ms100 request/s or 25 requests.
Since each server can handle 10 requests at the
maximum, we require a minimum of three servers in
the system.
This implies that on the average there will be 25
threads active in the system with 5 idle threads waiting
for new requests to arrive.
Little’s law Cont..
Example 2: Simple Corporation has decided to
implement Voice over IP
telephony on their intranet. The requirement is to
have no more than an
average of 4 requests pending at the IP-PBX
system, and the delay within
the PBX to be an average of 50 ms. Simple
Corporation expects an
average call volume of 500 calls/s. How many such
PBXes will they need?
Little’s law Cont..
Answer: Using Little’s Law, 4 equals the request rate times
50 ms, giving the rate for each PBX to be 80 calls/s.
For a call volume of 500 calls/s, they would need to have
nine systems.
Little’s law provides an estimate of the average properties
of a system.
Good planning principle dictates that one should not plan a
computer system to perform satisfactorily not just for the
average workload, but be capable of supporting the peak
workload with acceptable performance