Memory Overview

• This is still Jorgensen chapter 2.
  • But I am going to go in a different order.
• Main Memory
  • You can think of main memory as an array.
    • It has an address or index
    • At each address it has data.
  • Memory is generally byte addressable.
    • The data at location 0 consists of 8 bits.
    • The data at location 1 consists of 8 bits.
  • Frequently, however, we want more than one byte.
  • The term word is dependent on the computer
    • A word is the most general unit of operation for that specific computer.
    • Most computers today are 64 bit computers, or a word consists of 64 bits.
    • Many programming languages still operate with a word as a 32 bit element.
    • So a piece of data may be spread across multiple memory addresses.
    • Say we have a 32 bit number stored in memory.
      • It would require 4 bytes.
      • So it could be stored at location 0,1,2,3
      • Or it could be stored at location 32,33,34,35
    • In general the MSB would be at position 8 of the first address.
    • And the LSB would be stored at position 0 of the last address.
    • This is called little-endian.
    • We could also store it in the other way, which would be big-endian.
    • Different computers use different schemes
      • Bytes in order or in reverse order.
      • Bits in bytes in order or reverse order.
    • We need to be aware of this, mostly when sending data between machines.
  • In general however, programmers thank of memory as this array.
• But memory is more complex.
  • It is a problem of speed.
  • How do we measure speed?
    • In seconds, duh!
    • But that is not small enough.
    • 1 milisecond (1ms) is 1/1000 ($10^{-3}$)of a second.
    • 1 microsecond (1μs) is 1/1,000,000, ($10^{-6}$) of a second.
    • 1 nanosecond (1ns) is ($10^{-9}$) of a second.
    • 1 picosecond (1ps) is ($10^{-12}$) of a second.
    • Or there are 1,000,000,000,000 (1 trillion) pico seconds in a second)
  • The typical processor runs in the 100s of picoseconds per instruction.
    • Or a fraction of a nanosecond.
    • Or multiple billions instructions per second.
  • Registers must perform their work at the speed of the processor, so they operate in the picosecond range.
  • But memory operates in the 100 ns range.
    • Even fast memory
    • And there are other factors, such as bus transfer speed that impact this discussion.
  • So memory is at least 100 times slower than the processor.
• Cache is the way to mitigate this problem.
  • Cache is a layer between main memory and the registers.
  • Actually it might be multiple layers (L1, L2, ...)
• The operation of cache is somewhat complex, so we will delay discussion
• But cache is invisible to the programmer.
• Two pictures from your book:
  • 
  •