Why We Think GoLang Is Ready For Early Stage Startups(jellolabs.com) |
Why We Think GoLang Is Ready For Early Stage Startups(jellolabs.com) |
Modern Java is a lot less verbose than it used to be and it knows a lot of tricks (e.g. akka, play framework, dropwizard, Rx, dependency management, good IDEs, advanced profiling and instrumentation etc.)
Netflix is an excellent example of how you can leverage the JVM to build one of the most advanced system architectures hosted on AWS.
I'd say early stage startups are probably the least likely place for go to be a good choice - it sacrifices a lot for the sake of performance, which only matters once you've made it. As they say, scaling is a nice problem to have.
(Edit: and like Java it makes other sacrifices in the interests of being maintainable by less talented engineers. Again, that's something big companies need, not early stage startups)
You get both the performance, and you get a very lightweight language - I'd say almost as easy to write as Ruby or Python - with great open source support and growing and with a great community around it.
If you use Ruby or Python, you get extreme dynamism that makes things like ActiveRecord or dealing with external semi-structured data very easy (though you pay a performance cost).
If you use OCaml or Scala you get a lot of control over effects, making the code very safe and easy to reason about. And the type systems let you pull out high-level patterns in your code without going crazy, even when those patterns are very abstract, resulting in more concise and maintainable code.
From here go looks like the worst of both worlds.
Go development is fast, because the language is simple and it compiles very quickly.
Once you've made it, you certainly don't want to recode. Although, of course, if you've super-made it, you can just get more hardware. But even the largest web companies can run into the performance bottleneck. Like Google, apparently.
The Go Authors say one of the motives for building Go was the very long compile times of building C++ apps for large-scale web development. One thing they mentioned in particular was the large compile times due to processing and reprocessing of #include files. We know Microsoft ran into the same problem because they put Pre-Compiled Headers (PCH files) into their C family tools. Go has an import facility that prevents the problems of the #include file design.
Go is trying to hit the sweet spot for large-scale development. Go is by no means the only one, of course.
Any of you guys know how well .NET does in compiling big web-scale applications? I don't.
I don't dispute that, but those aren't the only competitors. If you're willing to give up the extreme dynamicism of those languages, you gain a lot more type-safety and expressiveness, and similar performance, from a modern strongly typed functional language like OCaml/F#/Scala/Haskell.
> Once you've made it, you certainly don't want to recode.
Yes and no. You probably want to accumulate a lot of technical debt during the early stages, meaning you're going to be doing substantial rewriting in any case. The rewrite strategy has worked out pretty well for e.g. Twitter.
> The Go Authors say one of the motives for building Go was the very long compile times of building C++ apps for large-scale web development. One thing they mentioned in particular was the large compile times due to processing and reprocessing of #include files. We know Microsoft ran into the same problem because they put Pre-Compiled Headers (PCH files) into their C family tools. Go has an import facility that prevents the problems of the #include file design.
Sure, like pretty much any language from the past 20 years. The C++ header recompilation problem is basically unique to C++. I can understand Go would seem good if C++ (and Ruby/Python et al) were the only alternatives, but that's really not the case.
I think the choice was much more of a proactive choice - we liked Go, we like the way it fitted us - good for fast iterative development upfront, good for the long term, fun to write code in and with a lot of advantages that come with the fact that it's a language designed today - with a lot of lessons learned from the past.
This is way diplomatic. Though for me your article has no beef, the Pros and Cons of Go has been repeated here for so many times. I was expecting to see the v.s. stuffs, why this, why that, your thought process and experiments. Right, those are the beefs I wanted from your article but didn't find.
>I think the choice was much more of a proactive choice - we liked Go, we like the way it fitted us - good for fast iterative development upfront, good for the long term, fun to write code in and with a lot of advantages that come with the fact that it's a language designed today - with a lot of lessons learned from the past.
This is not convincing though. As much as I admire Ken Thompson and Rob Pike, I tend to agree with the perspective that Go to C is like Plan9 to Unix. Of course, I could be wrong and am glad to be corrected by insightful comparisons and opinions.
Picking on Rails/Node was not my goal and I feared that a stronger comparison between these choices would have been seen as such by the current startup community.
At the end of the day most of these choices are tools and I truly believe that while they have their trade-offs, they are all good from one point or another. The point I was trying to make is that Go is a tool worth looking into as an alternative to Rails/Node for early stage.
Unless you build the exact same product in Rails/Node/Go it's hard to make meaningful broad comparisons - and you end up comparing lab experiments around speed in specific toy examples, or end up comparing only certain isolated parts of the toolset.
We are planning on writing more blog posts about our experience using it, more into performance internals and how it all plays out - but I do feel that we'll stay away from comparisons.