Why do we use Go?

At Reconfigure.io we’re creating a service which takes Go code, compiles it and deploys to FPGAs in the cloud. A major benefit of this is giving our users access to the hardware acceleration power of parallel computing without the usual hardware engineering requirements (VHDL, Verilog etc.). Parallel computing basically means performing many operations at the same time. Usually this is to process a huge amount of data, very quickly. FPGAs are made up of completely reprogrammable logic blocks that you can program to operate on multiple pieces of data, all in the same clock phase. While the parallel capability of a CPU is bound by the number of cores it has, FPGAs do not have this kind of limitation.

We could have chosen any popular high level language for the Reconfigure.io project, but Go has features that make it ideal for this task — It’s built for concurrency.

What isconcurrency?

A concurrent program is made up of well structured self-contained processes, which, if needed, could run at the same time. Concurrent programs translate well to parallel architecture for this reason. The self-contained processes can (loosely) be transposed to the FPGA logic blocks and therefore run in parallel.

Concurrent design is more complex than it might seem at first because generally we don’t want everything to happen simultaneously — where needed, we can design serialization into our concurrent processes. Take the example of image processing. It’s easy to see how the time taken to process an image could be reduced by breaking the image down into pieces and creating workers to process each piece, all at the same time (in parallel). But, once this has happened, the image needs to be put back together again. This step can only happen after the processing of each image segment has finished, therefore, serially.

Another feature of concurrent programming is that the resulting code has a higher degree of complexity compared to the non-concurrent version — more complex but potentially much faster. Take the image processing example from above, a non-concurrent solution would just involve one worker processing the whole image, pixel by pixel. But the concurrent program involves breaking the image up and passing each portion to one of several workers for processing, and then having the workers place the results somewhere so the final image can be pieced back together — a more complex yet more economical result.

How does Go help us achieve concurrency?

As well as being one of the most popular high-level languages, and a relatively easy one to pick-up, Go has a few key features that make expressing concurrency a simple process — areas a lot of languages leave to libraries.

Goroutines are lightweight threads that allow you to express syntactically where you want to introduce concurrency into your program. For instance, take this spin function that just prints numbers forever. If we were to call it like a normal function, it would run forever and subsequent statements would never run. However, if we run it with a go statement, a new lightweight thread is spawned that will run forever in the background while the main thread continues to execute:

func spin(){
for i := 0; ; i++ {
func main(){
// doesn’t block
go spin()


This isn’t a very interesting example, as it’s just a background thread, but we can use another feature of Go — channels — to communicate and synchronize across goroutines.

Channels are bounded blocking FIFOs that you can pass around just like any other value. There is syntax support for sending to and receiving from a channel. It’s pretty common in Go to use a channel where you would either share memory or use locks in another language.

This example spawns a goroutine to send to a channel, and then the caller waits to receive the value from the channel, and then prints it:

func send() <-chan int{
c := make(chan int)
go func(){ c <- 1 }
return c
func main(){
c := send()

Select statements allow the runtime to choose between multiple possible channel actions, either sending or receiving. Select statements implement ‘mutual exclusion’ which allows us to introduce serial aspects to our concurrent programs:

select {
case a := <-c1:
case b := <-c2:

All sounding familiar?

If you’re familiar with Go, but maybe new to parallel computing, are you comfortable working with goroutines, channels and select statements? Come and join in the discussion at our community page. Our service is being built for a great user experience and both feature development and supporting materials (tutorials, examples etc.) are guided by our community discussions.

If you’d like to read a bit more about this stuff, there’s some great talks by Rob Pike (Go creator): Concurrency is not Parallelism talks about how concurrency and parallelism are quite separate goals but provides some really simple, easy to follow examples about how concurrency can be achieved and how it can make parallelism easier. In Go concurrency patterns he talks through how various basic concurrency patterns can be achieved using Go.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.