Transcript Lecture 6

ICOM 6115 – Computer Networks
and the WWW
Manuel Rodriguez-Martinez, Ph.D.
Lecture 6
ICOM 6115
©Manuel Rodriguez-Martinez
Lecture Objectives
• Understand Network Programming
– Operating Systems support for Networking
• threads
– Implementation options for servers
• Iterative servers
• Concurrent servers
– Buffering of I/O streams
– Finite state machines
– Protocols Design
ICOM 6115
©Manuel Rodriguez-Martinez
Threads and Concurrent Servers
• One thread is running the master server
– Called the main thread
• Many other threads are used for slave
server
– Created on the fly
• MyServer2.java
– Obtained from pool of pre-allocated threads
• Will see later
ICOM 6115
©Manuel Rodriguez-Martinez
Processes and Threads
• Process
– Program in execution
– Has: program counter, text section, process stack,
and data section
– By default has one thread of execution
• Thread – lightweight process
– Unit of CPU time allocation
– Has: thread id, program counter, register set and
stack
– Allows for multiple actions to run within a process
ICOM 6115
©Manuel Rodriguez-Martinez
Multi-threaded Processes
Processes
Each thread is executing a part of the program
ICOM 6115
©Manuel Rodriguez-Martinez
Why threads?
• Responsiveness
– GUI can have one thread running HTTP client code
and other thread drawing JPEG file
• Resource sharing
– Thread shared variables in a process
• Sockets, and files can be shared by many threads
– Parallel I/O
• Simplicity
– IPC via shared memory is a pain …
• Support for parallelism on parallel machines for
free (“almost”)
ICOM 6115
©Manuel Rodriguez-Martinez
Example Program
• MyThread1.java
– Creates many threads that type the word
“Hello World” 10 times
• Java run time system calls the main()
method of the thread object
• Main() creates new threads and calls the
run() method in each one.
ICOM 6115
©Manuel Rodriguez-Martinez
Thread Types
• User threads (the ones you use)
– thread management done by User-Level
Threads Library
– Ex: Posix Pthreads, Solaris threads, Java
• Kernel threads
– thread management done by kernel of OS
– Ex: Windows 95/98/NT/XP, Solaris 2
• Need both to make multi-threaded
programs
ICOM 6115
©Manuel Rodriguez-Martinez
Thread Models
• Many-to-One Model
• One-to-One Model
• Many-to-Many Model
ICOM 6115
©Manuel Rodriguez-Martinez
Many-to-One Model
ICOM 6115
©Manuel Rodriguez-Martinez
Many-to-One Model
• Many user-level threads are mapped to a
single kernel thread
• Used on systems that do not support
kernel threads
– Sun OS 4
• Process will block on I/O request
• Cannot support parallelism on multiprocessors
ICOM 6115
©Manuel Rodriguez-Martinez
One-to-One Model
ICOM 6115
©Manuel Rodriguez-Martinez
One-to-One Model
• Each user thread is mapped to a kernel thread
• Provides more concurrency and paralellism.
• Process will not block for I/O
– Only the thread running the I/O request blocks
• Expensive to create kernel threads
– System bogs down when multiple applications are
opened
• Example OS:
– You guessed it! Windows 95/98/NT/XP
ICOM 6115
©Manuel Rodriguez-Martinez
Many-to-Many Model
thread dispatcher
ICOM 6115
©Manuel Rodriguez-Martinez
Many-to-Many Model
• User threads are multiplexed to multiple kernel
threads
• Provides more concurrency and paralellism.
• Process will not block for I/O
– Only the thread running the I/O request blocks
• Kernel threads can be pre-allocated
– Pool of threads
– Very inexpensive and fast
• Example OS
– Sun Solaris 2
ICOM 6115
©Manuel Rodriguez-Martinez
Problem: Shared Data
• Since threads can share many variables
we need a way to arbitrate access
• Lost update problem
– If many threads change a variable, which
change survives
– If many threads change a variable, how can
we serialize the changes to make them
consistent
• Withdraw from a ATM (aka ATH)
ICOM 6115
©Manuel Rodriguez-Martinez
Thread Contention
Threads
Balance - = 10;
Balance - = 20;
Balance = $400
ICOM 6115
©Manuel Rodriguez-Martinez
Balance - = 5;
Final Balance
Should be
$365
Solution: Synchronization
• Need a mechanism that guarantees only
one thread can access the variable at a
time
• Java provides synchronized methods
– Guaranteed to be executed by only thread at
a time
– Put code that changes shared variables on a
synchronized method
• Critical section
ICOM 6115
©Manuel Rodriguez-Martinez
Examples of Contention
• Example 1: Contention1.java
– Changes a shared variable unsafely
– Lost update problem
• Example 2: Contention2.java
– Changes a shared variable safely
• Using a synchronized method
ICOM 6115
©Manuel Rodriguez-Martinez