Whether using shared memory or distributed computing

Question

Go has the slogan "Do not communicate, exchanging memory, but share memory, exchanging." I was wondering if Go uses a shared memory or a distributed computing approach. For example, for MPI it is explicitly distributed; OpenMP is shared memory; but I was not sure that Go is unique.

I saw a lot of posts, such as Shared memory and Go channel linking , efficient Go document, etc., but could not clarify. Thank you in advance.

Answer 1

Go does not prevent memory exchanges between goroutines / threads. What they mean by exchanging data is that you send a piece of data or a pointer to the mentioned fragment on the channel. This effectively transfers ownership of the data to the target channel reader. Keep in mind that transfer of ownership is not provided by language or lead time, it is just by agreement.

You can still write to one memory from two goroutines if you want. In other words: Go doesn't stop you from shooting in the foot; it just provides language semantics that make these errors easier to detect.

If the value is passed to the channel, the programmer must then assume that the value is no longer its record in the same goroutine.

func F(c chan *T) { // Create/load some data. data := getSomeData() // Send data into the channel. c <- data // 'data' should now be considered out-of-bounds for the remainder of // this function. This is purely by convention, and is not enforced // anywhere. For example, the following is still valid Go code, but will // lead to problems. data.Field = 123 } 

Answer 2

The question suggests that shared memory and distributed computing are opposites. This is a bit like the question: do RAM and LAN have opposites? It would be clearer to distinguish between common concurrency memory in a CPU / memory node and between CPU / memory nodes.

This is part of a broader picture of parallel processing research. There have been many research projects, including:

• the development of computers other than Von-Neumann, which have several processors that use a single memory, to which some form of switching structure is attached (often the Clos network). OpenMP is great for them.

• development of parallel computers, consisting of a set of processors, each of which has its own separate memory, as well as some communication structure between nodes. This is usually the house of MPI.

The first case specializes in the brotherhood of high-performance computing. This is the last case that is familiar to most of us. In this case, usually these days, comms are simply via Ethernet, but for some niches, (successfully) faster alternatives with less delay have been developed (for example, IEEE1355 SpaceWire , which appeared from the serial channels of Transputer).

For many years, the dominant view was that effective parallelism would only be possible if the memory were shared, since the cost of transmitting a message by sending messages (naively) was considered prohibitive. With concurrency shared memory, the complexity lies in the software: since everything is interdependent, concurrency development becomes more complex and more complex as systems grow larger. Serious experience is needed.

For the rest of us, Go follows Erlang, Limbo and, of course, with Occam in promoting messaging as a means of choreography of the work to be done. This is due to the algebra "Transfer of sequential processes" , which provides the basis for creating parallel systems of any size. CSP designs are composite: each subsystem itself can be a component of a larger system without a theoretical limit.

Your question mentions OpenMP (shared memory) and MPI (messaging with distributed memory), which can be used together. Go can be considered approximately equivalent to MPI, as it promotes messaging. However, it also allows you to lock and share memory. Go is different from MPI and OpenMP, as it clearly does not apply to multiprocessor systems. To go into the world of parallel processing with Go, you need a network messaging infrastructure such as OpenCL , for which someone runs the Go API.