Hacker News Reader

Comment by wslh | Hacker News Reader

Having worked with some of the concurrency issues mentioned and observed both failures and successes, I’d like to share a perspective: developers who struggle most with concurrency often lack a formal Computer Science background. This isn’t to say that a CS degree is the only path to competence, but it provides structured exposure to concepts like concurrency through coursework, exams, and practical projects. These academic challenges can help solidify an understanding that’s harder to acquire in a fragmentary way

View on HN

While a formal Computer Science background is helpful, i don't think it is necessary for learning Concurrency. The reason people struggle with Concurrency is because they are not taught properly and also of course many are not willing to study and try and figure it out for themselves.

Once you start thinking about a program as a sequence of loads/stores (i.e. reads/writes to shared memory) and note that Pipelining/OOO/Superscalar are techniques to parallelize these (and other) instructions for a single thread of control you start getting an idea of how sequential order can be preserved though the actual execution is not quite so. Now add in another thread of control on another core and you get the idea of Memory Consistency problems and can be introduced to various "Consistency Models" (https://en.wikipedia.org/wiki/Consistency_model). Next introduce caches in between and you get "Cache Coherence" problems (https://en.wikipedia.org/wiki/Cache_coherence) Walk through a simple example like "c=a+b" in thread1 and "z=x+c" in thread2 and observe the interleavings and possible problems. Now move on to "Atomic Instructions" (https://en.wikipedia.org/wiki/Linearizability#Primitive_atom...) and how they can be used to implement "Locking" (https://en.wikipedia.org/wiki/Lock_(computer_science) ). Finally show "Memory Barriers" (https://en.wikipedia.org/wiki/Memory_barrier) as a means of enforcing a checkpoint on memory operations. All these together should make an understanding of Concurrency clearer. One can then jump back up into the language and study the language primitives for concurrency management and their common usage patterns.

I think that applies in general.

A formal education, in any domain, forces you to explore areas of knowledge that may not be directly applicable to immediate 'real world' problems. It does however give you the background to identify and explore those areas when you approach something that does overlap with work that people have spent literal lifetimes considering ahead of you.

I say this as someone without a formal CS education but has been working with computers for some time. I've spent countless hours, days, and months playing catch-up on that knowledge when I stumble into those blindspots.

Reply on news.ycombinator.com