ClientServer Model of Interaction

1
Client-Server Model of Interaction

• Chapter 20

2

• We have looked at the details of TCP/IP
• Protocols
• Router architecture
• Now want to look at application programs that use
  a TCP/IP internet
• Example applications are practical interesting
• Butfocus instead on the patterns of interaction
  among the cooperating applications
• Primary one is client-server paradigm
• Forms basis of most network communications
• Foundation for distributed algorithms
• See basic model here later chapters show use in
  specific applications

3
Client-Server Model

• Server
• Any program that offers a service that can be
  reached over a network
• Accepts a request
• Performs it service
• Returns the result to the requester
• For simple services, each request is a single IP
  datagram response is another datagram

4

• Client
• An executing program that sends a request to a
  server and waits for a response
• Client-server
• Extension of interprocess communication on single
  machine
• Servers can be simple or complex
• Time-of-day server returns current time
• Web-server fetches a copy of a web page

5

• Servers are usually application programs
• Execute on any machine supporting TCP/IP
• Timesharing system or a PC
• Multiple servers can offer the same service
• Can be on same machine or on multiple machines
• Copies on independent machines increase
  reliability
• Server can be applied to a machine
• When the computers primary purpose is to support
  a particular server program

6
Example UDP Echo Server

• Simplest form of client-server interaction uses
  unreliable datagram delivery
• UDP echo server
• Echo server process gets use of UDP echo port
• Enters infinite loop
• Wait for datagram
• Reverse source and destination addresses
• Return the datagram to the original sender
• Echo client is somewhere else
• Gets unused UDP port, sends UDP msg to echo
  server, waits

7
Figure 20.1
8

• Two points generally true of client-servers
• 1 - Operation
• Server starts execution before interaction begins
• Accepts requests and sends responses w/o
  terminating
• Client makes request and awaits response
• Terminates after using server a finite number of
  times
• 2 - Location
• Server waits for requests at well-know port
• Reserved for the service it offers
• Client allocates an arbitrary, unused,
  nonreserved port

9
Time and Date Service

• Time server
• Used to set a computers time-of-day clock
• Hardware device that keeps current date and time
• Can do manually
• Type in time and date when system boots
• Use client-server interaction to do automatically
• Configure one machine manually
• All others contact server when they boot

10

• Representation for the date and time
• Many store time and date from an epoch
• Use count of seconds since the epoch date
• UNIX OS uses January 1, 1970
• TCP/IP defines epoch as January 1, 1900
• Reports time as seconds past the epoch
• Time is 32-bit integer
• Representation is compact
• Allows easy comparison
• Ties together date and time of day
• Measure time by incrementing single binary integer

11

• Timer server interaction
• Interaction with time service like echo server
• Server sits at well known port
• Waits for UDP message
• Responds with UDP message containing 32-bit
  current time

12
The Complexity of Servers

• Sequential servers are simple
• Process one request at a time
• Assume OS queues the requests if server busy
• Queue will not get too long since work is trivial
• Usually more complex than that
• Must accommodate multiple concurrent requests
• Server typically has two parts
• Single master program to accept new requests
• Set of copies to handle individual requests

13

• Server performs following steps
• Open port
• Opens the well-known port where it can be reached
• Wait for client
• Start copy
• Invoke independent, concurrent copy of itself
• Copy handles the one request then terminates
• Continue
• Go to wait state while new copy handles request

14

• Processing requests proceeds concurrently
• New copy for each request
• Short requests finish earlier than long requests
• Independent of the order in which they are
  started
• Advantage of concurrent server
• Speed
• Disadvantage of concurrent server
• Complexity (more difficult to construct)

15

• Other forms of complexity arise as well
• Must enforce authorization protection rules
• Server programs execute with highest privilege
• Cannot blindly honor requests from other sites
• Must protect themselves against errors
• May get malformed requests
• May get requests that cause them to abort

16
Broadcasting a Request

• Maybe client will not know server address
• When booting, can use DHCP to obtain an address
• But does not know address of DHCP server
• For protocols where client does not know the
  address of a server, the client-server paradigm
  permits client to broadcast requests

17
Alternatives to Client-Server

• ARP protocol (Chap 5)
• Clients ask for information
• ARP minimizes such interactions
• Uses cache of answers to improve later queries
• Improves performance when recent history of
  queries is a good indicator of future use
• Does not change essence of client-server
  interaction
• Program executes until needs information
• Then becomes client to obtain the information
• Caching lowers the cost for all except the first
  process

18

• How to lower cost for the first request?
• In a distributed system, do precollection
• Background activities collect and propagate
  information before any program requests it
• Basis for UNIX ruptime command
• Reports CPU load and time since system startup
• Background program on each machine is used
• Periodically broadcasts information
• Collects incoming information and stores in a
  file
• Each machine always has a copy of latest info
• Client never has to access the network
• Reads info from secondary storage

19

• Chief advantage of precollection is speed
• Information available locally without delay
• Main disadvantage potential to be wasteful
• Uses processor time and network bandwidth
• Even when no one cares about the data being
  collected
• Cost for reading and processing broadcasts is
  high
• Precollection is not among most popular
  alternatives to client-server

20
Summary

• Distributed programs need communication
• Often use client-server interaction
• Server process awaits request then responds
• Client program makes request, sends, waits
• Client can send request directly or broadcast
• Broadcast useful on local network
• Machine not need to know address of server

21

• Servers can use internet protocols like UDP
• Accept and respond to requests across internet
• Can communicate with physical frames
• Restricted to single physical network
• Precollection is an alternative to the
  client-server model
• Precollects information to avoid delays