Operating Systems 2015F Lecture 24

Video

Video from the lecture given on December 4, 2015 is now available.

Notes

Lecture 24
----------

* /dev/random
* the Future

Randomness
 - unpredictable
 - unbiased (not predictable even in a statistical sense)

More precisely
 1. sequences of numbers that appear to be random
   according to statistical tests of randomness
 2. unpredictable

There exist sequences that pass #1 but not #2

pseudo-random numbers
 - digits of irrational numbers (e.g., pi)
 - almost any cryptographic primitive (encryption, hash)
 - pseudo-random number generators

How can computers ever be unpredictable?
 - generating pseudo-random numbers is easy, just
   run the right algorithm
 - using a random "seed" you can get arbitrarily long
   sequences of random-looking numbers that vary over
   time
 - but how do you get the seed?  And what about reversing
   the algorithm?

Computers cannot produce randomness on their own.
They need help from the outside world.

Sources of "real world" randomness observable by the OS
 - user input (keyboard, camera, microphone, mouse)
 - network traffic (inter-arrival time of packets)
 - device behavior (settling time of a spinning hard drive)

But *why* do you want randomness in a computer system?
* unpredictable program behavior
  - shuffling songs
  - games
* simulations
* randomized algorithms
  - genetic algorithms, other search algs
* cryptographic keys

For everything except cryptography, pseudo-random number
generators are more than good enough
 - caveat: some simulations need "high quality" random
   number generators

For crypto, you need things that are truly unpredictable

What sources of real world behavior are not
observable/potentially influenced by attackers?
* most can be observed or influenced

Solution: CPUs include hardware random number generators
* but, do you really trust them?

Instead, trust nobody but use everything you can

/dev/random & /dev/urandom
 - grabs whatever source of randomness it can, adds it to
   an "entropy pool", "stirring" it using
   cryptographically secure hash functions

/dev/random
 - only outputs as many bits as its estimate of entropy
   in the entropy pool

/dev/urandom
 - produces as many bits as requested; security is
   limited to number of bits in entropy pool

Use /dev/urandom for short-term keys, other stuff
Use /dev/random for long term keys

How good is /dev/random in a virtual machine?
 - depends on the hypervisor

The future
----------
An operating system turns the computer you have into the
computer you want to program (or use)
 - abstraction and resource management

OSs change when computers change and when how we use them
change

Embedded systems
  - many many kinds
Internet-of-Things
Personal devices
Tablets
Laptops
Workstations
Clusters
Data centers
Mainframes

Why in the world would one OS ever span all of these?
But we do!

Top-level concerns
 - network connectivity
    - heavy concurrency, certain device drivers
    - security
 - power constrained
    - in data centers, they care about
      "performance per watt"
      - because power and cooling are huge expenses
      
Feasible to use same code base because hardware is
all "powerful enough"

The future will be roughly the same, OS-wise

Change comes from duplication and refactoring

 - containers are just packaged OS-level virtualization

Learn the tools

New tools will come along to help handle complexity
 - but the concepts will be the same
 - and most of the mechanisms