Some OS and Networking Concepts

1
Some OS and Networking Concepts
2
Processor speeds

• Processor speeds (e.g. 2.6Ghz) are clock speeds
• An instruction consumes multiple clocks
• Different instructions take different of clock
  cycles
• Internal processor characteristics like
  pipelining (concurrent execution of instructions)
  make it even more difficult to just add up
  instruction exec times
• A simple way to get a rough estimate is to create
  a benchmark that has several thousand lines of
  executable code, and compute an average time per
  line of code
• Obviously a rather crude approximation
• Could easily be off by a factor of 2 or more

3
Virtual Memory and caches

• There are different types of memory with
  different access speeds
• Onboard high-speed memory has nanosec access
  times
• Matches speed of processor
• RAM has microsecond access times (1000 times
  slower)
• Multiple levels of caches in-between to bridge
  gap
• Caches store frequently/recently used data
  temporarily for quick access
• Like having multiple phone number/address lists
  city phonebook / websites, home computer, PDA,
  scribble on paper
• One more level beyond RAM hard disk, with
  millisecond access
• Program and data reside actually on disk, brought
  into memory and caches on need, written back out
  when space needed for something else
• Note that changes need to be propagated to lower
  levels
• Two design patterns write-through (immediate
  propagation to next level) write-back (wait
  till you are down and then make all changes)
• Time taken for memory access varies depending on
  how deep you have to go
• If same program is run again, time may vary
  because of cache differences

4
Processor registers
L1 cache (instruction, data) (few KB, lt 10ns)
L2 cache (instruction, data) (several KB, 100ns
access)
Copy out when done
Pull in on demand
(or propagate changes)
RAM (several MB, 1us access)
page3
page2
page6
page6
page8
page7
Pull in on demand
Copy out when done
Disk (several GB, 1-10ms access)
User Program
User Program
5
Program memory

• Code space the executable code
• Data space Globals, constants, statics
• Heap space Dynamic memory allocation
• Stack space Contains stack frames
• Each procedure call pushes another frame on stack
• Frame contains (space for) return address,
  parameters, local variables etc., as well as
  pointer to previous frame
• Can calculate memory budgets based on design
• Estimate LOC -gt code space
• Estimate total space needed for globals and
  statics
• Constants typically directly related to code size
  (estimate from history)
• Expected number of dynamically allocated objects
  and sizes -gt heap
• Stack depends on nesting depth of proc calls and
  size of local variables
• Memory still a scarce commodity for embedded
  devices PDAs, watches

6
Tasks

• Units of execution Includes code, data space,
  execution state, system resources file handles,
  network ports
• Two basic models, but actually a includes
  spectrum of possibilities in-between
• Processes Each process conceptually has its own
  copy of code, data, execution state and system
  resources
• Threads Conceptually, each thread has a separate
  execution state (line of control flow), but
  separate copies of all resources

7
Process1
Process2
Process3
Threads
Thread state
Memory
Memory
Memory
File handles
File handles
File handles
Socket handles
Socket handles
Socket handles
SharedMemory
Ports
Network
Files
8
Context Switching

• Processors run multiple threads / processes
  concurrently
• CPU runs one process/thread for a while, then
  another
• Different from parallelism multiple CPUs run
  several tasks simultaneously
• Switch between threads/processes when they get
  blocked
• Waiting for user to enter data, messages to
  arrive, resources to free up
• Blocked waiting on disk read to complete (disks
  much slower)
• Also switch if a thread/process runs for too long
• Otherwise if you are running a long
  compute-intensive program, others using same
  machine will just keep waiting
• Context switch Time taken to switch from one
  process/thread to another
• Involves saving/restoring registers state info,
  switching VM page tables etc.
• Thread context switch times are usually less than
  process context switch times
• Because threads share memory and other resources
  less changes needed
• Typical context switch times 10-100 microseconds

9
Scheduling

• Deciding which task (process/thread) to run next
• Two common schemes (there are many others)
• Priority scheduling Task with lower priorities
  only get to run if when all higher priority
  tasks blocked
• Round-robin Tasks take turns using processor,
  pulled off processor if when their quota (time
  slice) expires, wait for next turn
• Most common scheme is priority, with round-robin
  among same priority
• A pre-emptive priority scheme is one where the
  currently executing task is instantly pulled off
  processor if higher priority task arrives
• Needed in real-time applications to avoid
  priority inversion having higher priority
  task waiting on lower priority task
• Scheduling affects how long tasks wait before
  they get to run
• Needed for computing response times

10
Network concepts

• A network has
• A physical connection medium e.g. fiber-optic
  wire
• Characteristics bandwidth, transmission speed
• An access approach
• e.g. token ring (each node takes turns
  transmitting)
• contention (nodes transmit when they like and
  look for conflicts
• Characteristics average transmission delay,
  effective bandwidth, load behavior (e.g.
  contention behaves poorly at high loads)
• Protocols for transmitting data
• Message exchange sequences that coordinate
  transmission parameters, and ensure packets reach
  reliably (through acknowledgements)
• Characteristics reliability of transmission,
  packet sizes, networking overhead (some of the
  packets/space devoted to protocol stuff)

11
Networking concepts (contd)

• Approaches to handling networking concerns flow
  control, routing
• Characteristics End-to-end delays, fault
  tolerance
• Data processing Breaking up and recombining
  packets, (possibly) encryption and compression
• Characteristics Processing time, networking
  overhead
• Networking involves multiple hardware/software
  layers
• ISO-OSI 7-layer conceptual model of network
  layers

12
TCP/IP, Sockets and Ports

• A combination of two protocols TCP and IP
• Provides reliable end-to-end transmission
• A port is a logical address on a node. Program
  associate themselves with ports for input/output
• A socket is a program-level data structure for
  accessing a port, like file handles
• Higher-level concepts like RMI are built on top
  of TCP/IP
• TCP/IP (and most other protocols, but not all)
  provide a byte-stream view of what is being sent
  received