Where the time actually goes.
WHAT CACHE IS
Small fast memory holding recently and likely-used data.
WHAT LOCALITY MEANS
Temporal: recently used data is likely used again
Spatial: data near recently used data is likely used
WHY IT MATTERS
Cache design assumes both, and programs violating them perform badly.
WHAT A CACHE LINE IS
The unit transferred, typically several dozen bytes.
WHAT THAT MEANS
Accessing one byte fetches its neighbours.
WHAT SEQUENTIAL ACCESS ACHIEVES
Every fetched line fully used.
WHAT RANDOM ACCESS ACHIEVES
Most of each fetched line wasted.
WHAT THE PERFORMANCE DIFFERENCE IS
Substantial, frequently an order of magnitude.
WHAT FALSE SHARING IS
Threads writing different variables that share a cache line.
WHY IT IS DAMAGING
The line moves between processors constantly, destroying performance with no apparent cause.
WHAT VIRTUAL MEMORY PROVIDES
Each process its own address space, mapped to physical memory.
WHAT PAGE FAULTS ARE
Accessing memory not currently resident, requiring it to be fetched.
WHAT SWAPPING CAUSES
Catastrophic slowdown, because storage is orders of magnitude slower than memory.
WHAT TO MEASURE
Cache misses, not only instruction counts.
