CMPT 225 Lecture 11: Lecture 11
25 Jan 2019
School
Department
Course
Professor
Document Summary
Program running time depends on algorithm fundamentals, hardware speed, quality of programming. Our standard model of hardware has cpu connected to i/o to disk, network, user. In theory arithmetic operations take time 1 and so does accessing data in ram. But in reality operations in cpu about 200 times faster than to access an int in ram. Modern dram moves data fast: about as fast as a cpu can process it. Latency is how long from request to delivery. Trans alaska pipeline takes 4 days to redirect oil. Modern dram takes as long as 200 in operations. An unpleasant fact: for many applications, execution time is dominated by memory latency. Thus we created cpu cache which is close to the cpu. The cpu requests a byte from memory, system first checks if that byte is in cache, if it is ~2 20 cycle latency. If no ~200 cycle latency to read from memory.
