Computer System Lecture Notes
Class 3 Hour 1
                    -- by Guanghao Ding(ding.g@husky.neu.edu)
==============================================================

== Tonight's Topics
    * Threads / Threads Scheduling
    * Scheduling Issues
    * Syncronization

**************************************
*   Threads / Threads Scheduling     *
**************************************

== Motivation
What's the limitation of processes, i.e. why threads are useful?
* context switching is expensive
* sharing data is cumbersome
* applications can only do one thing at a time
* high memory overhead

== What we have to copy during context switching
1. stack, heap are not necessary 
2. registers 


So we can have a solution like this 

..............................     ...............................
| text       data        I/O |     | text      data         I/O  |
|----------------------------|     |-----------------------------|
|  register            stack |     |registers|registers|registers| 
|----------------------------| ==> |---------|---------|---------|  
|                            |     | stack   | stack   | stack   |
|                            |     |---------|---------|---------|
|          threads           |     | thread  | thread  | thread  |
|                            |     |         |         |         |
``````````````````````````````     ```````````````````````````````

: it helps with "high memory overhead", the context switching is a little bit 
  cheaper, but doesn't help data sharing. 


== Benefits
1. Responsiveness
    e.g. email client: one thread waiting for server, 
            another thread waiting for input and so forth.
2. Resource Sharing


== Challenges 
    : sometimes it's hard to divide the task into multi parts,  
    not that simple to be managed. e.g. How to make maximum 
    flow problem multi-thread? The simplest answer is that  
    write separate thread to find path, when the sub-threads get 
    back, the main thread decides which path to remove. 

== Implementing Threads - Multithreading Models

* One-to-One Model
	AD(advantage):  
* Many-to-One Model
    : kernel doesn't know the multi-threading in user space, 
      only one application
    ADVANTAGE: we can build a better scheduler since user
        know what the threads are doing, and context switching
        is cheap since kernel space switching is avoided.
    DISADVANTAGE: poor blocking behavior, one block, 
        all blocked.
* Many-to-Many Model
    : get the benefits of the above two, can be operated as 
      many-to-one and one-to-one


== Implementing Threads - Basic Functions
* create
* cancel
* join

== Threading Issues
* fork()
* SIGNAL
    : notify a process that a particular event has occurred. 
    if no one interests in the SIGNAL, just ignore or pick one randomly
* Threads Pool
    e.g. Web Server. If the server create a new thread for every request,
    the server can be easily brought down by sending huge amount of requests. 
    Threads pool are useful in this case.