Review: Process Management

1
Review Process Management

• Objective
• Enable fair multi-user, multiprocess computing on
  limited physical resources
• Security and efficiency
• Process running program context
• Process Control Block Contains Process Context
• Register content
• Memory address space
• PC
• I/O Status

2
Review Process Management

• Multicomputing is achieved by timeslicing Every
  process runs for a while until time is up
• Interrupt driven(2 Types), PCB is stored and
  retrieved
• 5 State model new, running, ready,
  waiting,terminated
• multiple queues in which processes reside during
  lifetime

3
Diagram of Process States
exit
new
terminated
admitted
interrupt
Queue ready
running
scheduler
I/O or event wait
I/O or event completion
Queue waiting
4
Review UNIX Commands

• telnet machine login to UNIX computer
• ps show all processes that were started
• ps -u username all processes started by user
• top show all processes running on the computer
• kill id Stop running process with id
• nohup make process immune to high level
  interrupts
• nice int process change the priority of a
  process
• man command get information about command

5
Inter Process Communication

• Passing information sequentially between 2
  processes (simplest case)
• Result of P1 becomes at termination the input to
  P2
• Passing information concurrently between 2
  running processes
• Use of shared resources (such as printer queue)

6
Tools for sequential communication

• Redirect
• gt saves the output (whatever would come to the
  screen) into a file
• gtgt appends the output to a file
• Example print gttmp
• Pipe
• hands the output directly to the next process

7
Example

• See the 9823th line of a file
• head -n file command that shows the first n
  lines of file
• head -n file command that shows the last n lines
  of file
• 1) Store the first 9823 temporarily in a file and
  than look at the last line
• head -9823 filegttmp
• tail -1 tmp
• 2)Pass results directly using pipe
• head -9823 filetail -1

8
Communication between processes Sharing

• Simplest method any 2 processes can use shared
  file that both processes can access
• However, only one process at a time can have
  write access
• Parent-child processes can have shared variables
• a C program can start of new processes (fork)
  that have access to all global variables

9
Use of shared resources

• Example 2 unrelated processes want to print to
  the same printer
• Printer collects jobs in a printer queue
• (lpq -Pprinter shows you all entries in the
  queue)
• A queue works like an array. Each entry is a
  file that has to be printed. Processes can submit
  a file by appending it to the array. The printer
  takes jobs out from the lower end of the array.
• Global variable LAST keeps track of end of the
  array

10
Problem of Sharing Inconsistency

• Multiprogramming can lead to inconsistency in
  case of shared resources
• P1 and P2 want to print, P1 is running and P2 is
  next in the ready queue
• P1 loads the current value of LAST into register
  R1
• P1 gets interrupted by scheduler because time is
  up
• P2 starts running and loads LAST into R3
• P2 stores filename in array(R3), increments R3
  and stores R3 into LAST

11
Inconsistency II

• P2 gets interrupted by scheduler and P1 starts
  running (remember, OS will restore all register
  values)
• P1 will store its filename at array(R1) believing
  that R1 is the current value of LAST and
  overwrite the filename stored by P2
• P1 increments R1 and saves the value to LAST
• P2s file will never be printed!

12
Problems of Sharing Deadlock

• Deadlock Processes are waiting for resources
  held by another

• P1
• lock(x)
• lock(y)

• P2
• lock(y)
• lock(x)

• P1 cant finish since it is waiting for resource
  y
• P2 cant finish since it is waiting for resource
  x
• Neither of both ever unlocks the one it has

13
Race Conditions

• Whenever sharing of resources can lead to
  inconsistencies or deadlock
• Problem Very hard to debug, not reproducible
• Heart of the problem is that between a test/load
  of a resource and the update there could be an
  interrupt
• Solution atomic instruction or uninterruptable
  parts of code

14
The Dining Philosophers
Plate with food
Fork
15
Philosophers

• 5 Philosophers either think or eat
• If a philosopher gets hungry he tries
  sequentially to obtain 2 forks, one from his
  right, one from the left
• How do you make sure that they dont starve?

16
Solutions 1 Test and Set

• Many processors offer a TSL instruction
• TSL R1, LOCKLAST
• It reads the value and during the same
  instruction sets it to 1.
• Now there cant be an interrupt between the load
  and the store as we had in the printer queue
• If R1 is 1, the process will wait, if it is 0 it
  means that the resource is available and now
  locked by P1
• This simple test can be used before altering LAST

17
Solution 2 Semaphores

• The concept of semaphores extends the idea of
  test and set to an integer value.
• A semaphore is a variable that can be incremented
  or decremented without interrupt

18
Solution 3 Critical Block

• If multiple resources are needed as in the
  deadlock case, we have to ensure, that either all
  resources are allocated or if only one missing
  none of them.
• This operation of locking all has to be immune to
  interrupts
• A critical block is a clock of instruction during
  which a program is immune to interrupts
• Note this is a dangerous game and you dont want
  to leave this to the user to implement! What
  happens if it contains a infinite loop?

19
Solution 4 Detect and Recover

• All previous attempts aimed at prevention
• Sometimes it is easier and safer to allow for the
  possibility of deadlocks
• Deadlock detection Dependency graph of resources
  has cycles
• Graph for example
• In case of deadlock, OS prevents second process
  to allocate resource that produces a cycle

20
Summary Solutions

• Inconsistency can be avoided by ensuring
  uninterrupted blocks of instructions
• Solutions might involve hardware support (if
  there is a test and set instruction that is
  supported by the CPU design)
• Software support can avoid an interrupt by
  changing the responsiveness to interrupts in
  critical sections
• Software can detect deadlock cases and resolve
  them

21
Appendix More UNIX

• more file show content of file
• ls show all files in that directory
• head -n file Show first n lines of file
• tail -n file Show last n lines of file
• gt redirect output into new file
• gtgt append redirected output to new file
• pipe output into process