This is probably one of the most intense books I’ve ever read. It may be intense because I intended to power through it once in one week and go through it later in a series on concurrency models on Bytes by Ying rather than methodically read it in seven weeks (where I might forget to continue to read this book or not want to read it anymore).
Multicore programming was the future twenty years ago, so naturally its time is now. I genuinely think, unlike artificial intelligence or deep learning, a mastery in distributed programming and concurrent/parallel programming will prove to be valuable knowledge in many areas of expertise, for the sole reason that you really can’t find a single-core processor in the professional/enterprise anywhere anymore. This has played out in the business field as well; efficiency gains from using Erlang in production allowed WhatsApp to sell to Facebook for nineteen billion dollars with only fifty-five employees. Being able to service something like a hundred million people per employee, with nine nines of availability without patches, on a handful of servers, allows you to do that. Being efficient and smart in your allocation of resources in engineering just gives you so much more breathing room in business, and using a great concurrency model allows you to be both efficient and smart.
Butcher does a great job of describing the underlying hardware restrictions and how they influence each concurrency model and where it should be used. He starts off with threads and locks, and how much mental overhead it takes in order to use that model effectively. He then describes alternative concurrency models and how to apply them within a language, under what requirements and constraints they are effective, and at what scale they become worthwhile.
One thing I wish he had covered more was how your usage of dependencies would impact your concurrency model; how friendly is the development and developer ecosystem to the concurrency model you wish to leverage? Another thing was how idiot-proof is each concurrency – that is, if you made an extremely bad hire and somebody was screwing up in your codebase, how would that impact your concurrency model?
Ultimately, it boils down to this: using the right concurrency model for the right problem will dramatically improve release cycles for performant software, which will make your business and your objectives that much more attractive and reachable.
This is absolutely a must read for any serious software engineer.