202309061217
Status: #idea
Tags: OS

Multi-processor Model

Multi-core vs Multi-processing

Multi-core

• L2/L3 cache is shared between cores
• Can have upto ~100 cores

Multi-processor

• RAM sharing
• Can be via motherboard, or via network (in the case of distributed computing)
• Can scale to even 1000x processors
• 2 types
  1. Asymmetric
     • Each processor is assigned a specific task
     • Example: A co-processor
  2. Symmetric
     • Local registers and cache are separate, but share main memory
     • Can even be at ISA level like big.LITTLE (P-cores & E-cores)

Multi-processor Memory Model

Uniform Memory Access (UMA)

• Physically impossible
• All levels of memory have equal access time

Non-Uniform Memory Access (NUMA)

• Getting data from different levels of memory takes different amount of time
• Accessing my own L1 is faster than another core's L1
  

NO Remote Memory Access (NORMA)

• Can not access another core's data directly

Procrastination

• Putting a job off, to increase shutdown time and save power
• Example
  J1: 0->10 (1 EU)
  J2: 10->30 (2 EU)
• Execute J1 from 0->1. Shutdown till 28. Execute J2 from 28->30

References