Question

1. Briefly explain what is The von Newmann bottleneck and describe at least one solution to resolve this bottleneck problem. (10 points)

Answer 1

Definition:

A systems bottleneck, in that the bandwidth between Central Processing Units and Random-Access Memory is much lower than the speed at which a typical CPU can process data internally.

The Von Neumann bottleneck is the idea that computer system throughput is limited due to the relative ability of processors compared to top rates of data transfer. According to this description of computer architecture, a processor is idle for a certain amount of time while memory is accessed.

Solutions are,

1)Cache:

Caches help to mitigate this problem for many workloads, by keeping a small amount of frequently used data close to the processor. If your workload reuses a lot of data, as many do, then you'll benefit from caching. However, if you are processing a data set that's too big to fit in cache, or if your algorithm doesn't have good data reuse, it may not benefit much from caching. Think processing a very large data set. You need to load all the data in and store it back out at least once. If you're lucky, your algorithm will only need to see each chunk of data once, and any reused values will stay in cache. If not, you may end up gong over the memory bus much more than once per data element.

2)Parallel processing:

Shared Memory

The way shared memory processors are implemented today doesn't do much at all to solve the Von Neumann bottleneck. If anything, having more cores puts more strain on the bus, because now more processors need to fetch data from memory. You'll need more bandwidth to feed all of them. Case in point: many parallel algorithms are memory-bound, and they can't make use of all the cores on modern multi-core chips, specifically because they can't fetch data fast enough. Core counts are increasing, and the bandwidth per core will likely decrease in the limit, even if the total bandwidth increases from processor to processor.