Go vs. Rust: Productivity vs. Performance (2014)

http://joshitech.blogspot.com/2014/11/go-vs-rust-productivity-vs-performance.html

by daviducolo • 11 years ago

52 64 11 years ago

joshitech.blogspot.com

Great article, and I have to say I completely agree with your point on using go for enterprise level development.

kibwen16• 11 years ago

Taking an average of wildly varying microbenchmark scores doesn't seem particularly useful. Furthermore Rust has never had particularly optimal implementations for the benchmarks game (I can't speak as to the quality of the Go implementations). Furtherfurthermore this blog post is so old that the Rust results cannot be considered representative. Furtherfurtherfurthermore I would like it if we could just all collectively stop treating the benchmarks game as though any of its microbenchmarks measure anything other than "is this language capable of binding to libpcre" or "is this language capable of binding to libgmp". I swear, the benchmarks game is the SunSpider of programming language comparisons.

(Furtherfurtherfutherfurthermore, why are we still comparing Rust to Go? I thought we had gotten over this!)

lucozade• 11 years ago

Although you are of course correct, I would caution against a complete dismissal of micro benchmarks.

Firstly because they are a sort of marketing: I've perused the shootout before and noticed an implementation get a higher score than I was expecting. This has sparked an interest in digging deeper into that implementation/language.

Secondly because of the "can't even" problem i.e. why should I invest time investigating a new language if it can't even do simple benchmarks well? I appreciate that this isn't a particularly analytical view but I expect that it's not uncommon.

kibwen16• 11 years ago

It's a mistake to consider the benchmarks game a "can't even" problem, the level of effort given to each language's implementations is all over the map.

This also isn't an utter dismissal of the concept of microbenchmarks, it's an utter dismissal of the concept of microbenchmarks as used to compare language implementations. Something like the TechEmpower web framework benchmarks is at least marginally better because the tasks permit a wider variation of implementations and at least pretend to do some kind of useful tasks, instead of pretending to be an arbiter of implementation quality (which, as much as the benchmarks game disclaims it, is still what people take it as).

igouy3• 11 years ago

>Taking an average of wildly varying microbenchmark scores doesn't seem particularly useful.<<

Correct. Which is why, unlike Rohit Joshi, the benchmarks game simply shows a broad median; and when a summary measure is required to order the box plots, uses the geometric mean:

http://benchmarksgame.alioth.debian.org/u64q/which-programs-...

(Note the reminders -- "Please don't obsess over tiny differences in median values from such a small number of examples." -- "Please don't obsess about which programming language implementation is shown 10th and which is shown 11th. You can see that the order would be different if it was based on the median scores instead of the geometric mean scores.")

>as though any of its microbenchmarks measure anything other than "is this language capable of binding to libpcre" or "is this language capable of binding to libgmp"<<

Incorrect. Everyone can see that there is only 1 regex task and only 1 arbitrary precision arithmetic task -- and they sensibly allow library use.

honest_joe4• 11 years ago

Rust's not there yet and I wonder if it can compete. The thing is it is too specialized and low level (not that it is a bad thing for what it is intended for). The problem with all of the low-level things is that they are seriously undervalued.

I am glad that there is servo but we need something like an operating system (IoT might work for the start). I hope Samsung changes its plans with Tizen and eventually will develop something with Rust in mind.

kibwen16• 11 years ago

I wouldn't be worried about Rust competing, just because HN revolves around web tech doesn't mean the rest of the world undervalues systems programming. :)

honest_joe4• 11 years ago

I meant financially undervalued by companies. Yes you have some players here and there but overall the market is not that big. And an investment to this area from a developer's POV is too much of a risk. Also the learning curve is too steep.

Web tech is just the simplest way to reach the end user. Mobile apps or desktop apps are becoming a hassle as well. It just takes time to "install" something and use it.

I hope thing change but I am skeptical.

Jweb_Guru8• 11 years ago

I have heard a couple of comments about Rust not being "there" yet. Why is it not "there" for you? Where, precisely, is "there"? I certainly hope you are not basing this opinion on a blog post from 2014 with terrible methodology (even the maintainer of the benchmarks came down to comment on how egregiously he was misusing the numbers).

honest_joe4• 11 years ago

Lets put it this way. Rust is great and all for somebody who is willing to invest in it but there are simply other "mature" alternatives that are just as good.

pjmlp5• 11 years ago

Not take this as a post against Rust, I like the language and what is trying to be achieved.

Just listing some of the issues, from my point of view, Rust currently faces.

It is not there if one is looking for uses cases where GC performance doesn't matter e.g. Haskell, OCaml, .NET, JVM. All existing alternatives enjoy more mature libraries, editors, supported OSes and so on.

It is not there if the goal is to replace C++ on the mobile platforms, as it introduces hurdles and increases the development time versus using the vendor tooling.

Specially in the iOS and Windows Phone, where C++ is a first class language. A kind of Objective-Rust, Rust/CX are needed there, even if rustc is able to cross compile.

On Mac OS X it is not there in regards to Swift tooling and available libraries, additionally for many developers ARC and GDC might just be good enough, even if Swift is less safe than Rust.

noelwelsh• 11 years ago

Every big data platform I know of is or already has implemented manual memory management (they all run on the JVM). I can see Rust doing well in this world. It's hard to say things like Spark are undervalued. (Two data points: ~$50M investment in Databricks, the company commercialising Spark. IBM just committed ~3.5K developers to working on Spark projects.)

kibwen16• 11 years ago

Big data could be huge for Rust. I have a friend who works in the Hadoop ecosystem who's about to take a 90% pay cut to leave his current job in favor of using Rust at a startup. If I had gratuitous amounts of money, this is the field that I'd be investing in.

pella• 11 years ago

The new Cockroachdb implementing in GOlang https://github.com/cockroachdb/cockroach

anhtran• 11 years ago

I think Rust is a fledgling this time. But I love the clearly orientation that they focus on. It is a safety system language, especially, the ownership/borrowing.

pjmlp5• 11 years ago

Go seems to be for the Python, Ruby and Java developers and will be used for enterprise applications, mobile apps and application servers.

No thanks, Go offers no benefit over the Java eco-system, neither in terms of deployment[0][1][2], not in terms of libraries and dependency management, nor in language expressiveness (Java 8, Scala, Clojure, Kotlin, Ceylon, Groovy)

[0] http://www.excelsiorjet.com/

[1] http://robovm.com/

[2] http://www-01.ibm.com/support/knowledgecenter/api/content/SS...

EDIT: Also forgot to add that we enterprise devs also have C#, F#, Swift, C++14, C++/CX at our disposal

copx• 11 years ago

[0] Price per developer (!) $3000, only supports Java 7. To compare, the price of the Go compiler per developer: $0

I see one clear benefit right there.

pjmlp5• 11 years ago

I just wanted to list a few examples.

Almost all commercial JVMs do support AOT compilation.

Plus we are talking about the enterprise here, those values are peanuts in project budgets.

Also don't forget to add "developer time * cost per hour" writing repetitive code in Go to your price.

dleskov2• 11 years ago

Low/zero cost is a benefit only if everything else is equal. Eclipse and Netbeans are free, yet people pay for IntelliJ IDEA. We use an expensive commercial GUI testing product because all the free and cheaper alternatives sucked big time, and support was either poor or non-existent.

dleskov2• 11 years ago

[0] can be had at $10+ under a charity program at the moment: http://www.excelsiorjet.com/charity

ExpiredLink• 11 years ago

I think Rust will attract developers from C, C++, Fortran and will be used for developing high performance systems like gaming, browsers, telco servers, distributed computing systems as well as low level, cpu efficient embedded/micro computers.

People choose platforms, stacks, languages, environments, not languages: Rails not Ruby, browser not JavaScript, Unix not C, host not Cobol, ...

unfamiliar• 11 years ago

Fortran programmers will never move to Rust. The level of systems knowledge required to use Fortran is almost nothing. It is already a pretty safe language due to the heavy restrictions on pointers and compiler warnings. The abstractions and restrictions in Rust will make it completely unpalatable to the scientific community.

pjmlp5• 11 years ago

If anything they might eventually adopt Chapel, specially it is has the sponsoring of a few CPU vendors.

Also Fortran is getting a new standard revision this year.

kibwen16• 11 years ago

I'm not sure that Rust particularly cares about attracting Fortran programmers. Fortran is a highly specialized language whose most powerful asset is its enormous body of working code. That's not a market that needs another competitor.

kele• 11 years ago

The article has a major flaw: it confuses productivity with LoC.

nickysielicki• 11 years ago

they don't exist in the same space. Go is not a systems programming language and Rust is not a scripting language.

alediaferia• 11 years ago

Despite being always interested in seeing benchmarks comparing new languages I don't feel like basing my choices on such benchmarks. From my experience it's usually a matter of how comprehensive is the ecosystem surrounding the language. Anyway, great article.

jpgvm• 11 years ago

This article doesn't really match my experiences.

I have found Rust to be considerably more expressive than Go. Traits and proper generic types, iterators and pattern matching all contribute to this. I tend to write less lines of Rust code to accomplish the same tasks.

As for runtime performance, in my experience they are pretty close right now for my use cases but I do expect Rust to grow faster than Go over time.

collyw• 11 years ago

"Go seems to be for the Python, Ruby and Java developers and will be used for enterprise applications, mobile apps and application servers."

Out of interest, what would I gain by learning GO? I already know Python pretty well. I have done Java in the past and don't see any problem jumping back into it. Both languages have tons of libraries and tooling available. Are mature, and I can't see much reason to opt for a new language.

falcolas• 11 years ago

what would I gain by learning GO? I already know Python pretty well.

A significantly easier deploy story. Compile once on your target platform (i.e. Ubuntu), run on any of the same platform, without having to worry about installing (or how to install) all of your third party dependencies.

It's also more friendly to less skilled developers, with its type system and limited metaprogramming.

It also has a really compelling concurrent programming paradigm which doesn't involve libraries or pickle.

Of course, those same strengths are also weaknesses - the language is less expressive and flexible, and it requires extra steps to compile and distribute those compiled binaries.

flippinburgers2• 11 years ago

Go is a simpler language and as such it is a better language.

EugeneOZ2• 11 years ago

Then BASIC is most awesome language.

flippinburgers2• 11 years ago

Always use the simplest solution that solves the problem. I didn't realize that BASIC has concurrency. /s

Go vs. Rust: Productivity vs. Performance (2014)

Related Stories