Douglas6 wrote: ↑
Sat Jan 20, 2018 10:04 am
These posts always make me wonder if car manufacturers get customer support questions like "Can I put two Ford Fiestas together to get a 300 horsepower car?"
It is certainly possible to hitch horses and then count the number of horses in the team. Apparently the right kind of horses sense what the others are doing and they all pull together. Although locomotives can also be hitched together to pull a train, Ford Fiestas were engineered according to more restrictive constraints and can not be hitched into a team--the same with chickens. Quoting Seymour Cray
Seymour Cray wrote:If you were plowing a field, which would you rather use: two strong oxen or 1,024 chickens?
The question then becomes whether a Raspberry Pi is more like a chicken or an ox. Is one Pi able to sense what another Pi is doing and compute in the same direction?
Just as oxen and chickens have eyes and ears, a Pi requires some sort of communications hardware such as a network connection to sense what the other Pi is doing. The built-in 100-Mbit Ethernet of the Pi has twice the latency and 1/10 the bandwidth of a typical desktop computer. It is also a factor of 1000 slower than the interconnects used in typical computing clusters. Imagine driving a team of partially bind horses that can't hear.
On the other hand, chickens can see and hear fine, but generally have no desire to pull in the same direction. This is likely a software issue that is sometimes described by the phrase bird brain. Likewise, most software people run on one Raspberry Pi doesn't even try to leverage the help of another Raspberry Pi to increase performance.
Having said all of this, it is generally agreed that increasing parallelism either through quantum computing, GPUs, multiple cores or clustering is the way forward from a performance point of view. Lots of people are performing tasks that are trivial in the sense that they fit within the limitations of a single traditional processor thread. At the same time, with the exception of quantum parallelism, the Raspberry Pi provides a cheap but effective environment for learning about parallel computing.