HTTP server running in parallel via Isolate

[Michele Costa]

Hi,

i am the maintainer of a micro-benchmark branch for several different languages: i test a plain "Hello World" HTTP server by using the `wrk` loading tool.

I've recently added Dart to the benchmark, by relying on the async HTTP server implementation i've found on official tutorials.

The resulting throughput is not disappointing, but nonetheless far from the Node.js one, which uses  the cluster library to pre-fork several processes.

Dart HTTP server implementation does not run in parallel, as confirmed by inspecting the CPUs usage via Activity Monitor.

I read that Dart deliver parallelism on a multi-core server via the Isolate library.

Unfortunately i have found no concrete examples of how to use Isolate on the standard APIs and the code-snippets i found online are not relevant and/or obsolete.

Any suggestions on how to implement a HTTP server by using Isolate.spawn will be really appreciated.

Many Thanks

Mike 

[Kevin Moore]

Here you go! Using pkg/shelf – but you could do your own thing. Setting `shared` to true is the key bit.

import 'dart:io';
import 'dart:isolate';
import 'package:shelf/shelf.dart' as shelf;
import 'package:shelf/shelf_io.dart' as io;

shelf.Response _echoRequest(shelf.Request request) =>
    new shelf.Response.ok('Request for "${request.url}"');

main() async {
  var isolates = 4;

  for (int i = 1; i < isolates; i++) {
    Isolate.spawn(_startServer, []);
  }
  _startServer();
}

_startServer([args]) async {
  var server =
      await HttpServer.bind(InternetAddress.LOOPBACK_IP_V4, 8080, shared: true);
  io.serveRequests(server, _echoRequest);
}

[Zach Anderson]

There's no need to use shelf. https://github.com/costajob/app-servers/pull/23

--
For other discussions, see https://groups.google.com/a/dartlang.org/
 
For HOWTO questions, visit http://stackoverflow.com/tags/dart
 
To file a bug report or feature request, go to http://www.dartbug.com/new
---
You received this message because you are subscribed to the Google Groups "Dart Misc" group.
To unsubscribe from this group and stop receiving emails from it, send an email to misc+uns...@dartlang.org.

[Kevin Moore]

On Tue, Oct 3, 2017 at 12:21 PM, 'Zach Anderson' via Dart Misc <mi...@dartlang.org> wrote:

There's no need to use shelf. https://github.com/costajob/app-servers/pull/23

Absolutely true! Sorry. I'm biased: https://github.com/dart-lang/shelf/commit/8fd1b3f1ac920cdf67a34364ef8593e4a46af1c3

 

You received this message because you are subscribed to a topic in the Google Groups "Dart Misc" group.
To unsubscribe from this topic, visit https://groups.google.com/a/dartlang.org/d/topic/misc/Ju4f2d5ziwE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to misc+uns...@dartlang.org.

[Michele Costa]

Hi all,

thanks for your support: i was now able to test versus a parallel server.

I am not aware it is so easy using the Isolate library.

Throughput is now better and distributed on available CPUs.

Best

Mike 

[Filip Hracek]

Hi Mike, I'd be interested to see how this impacted the results. Will you be making an update to the table in README? 

[Michele Costa]

Hi Filip,

i've updated the README: throughput using all the cores is now on par with Ruby, about 27k reqs/sec.

Memory consumption was the same as before, but CPUs usage is now 240% (before was 110%).

Numbers testing the server with WRK on localhost are higher, but so are for all of the others languages.

The fact that Dart can use threads to deliver parallelism is a win-win to me, compared to the pre-forking model of Node, Ruby or Python, all faster to be fair...

[Patrice Chalin]

Hi Mike: the README refers to version 1.2.4 of Dart. I suspect that you probably meant 1.24.2, right? --Patrice

[Michele Costa]

Right! Just changed that, thanks!

[Joe Conway]

There is still a lot left on the table here. On a four core machine, you can use 5 isolates before you get diminishing returns (here is a graph: https://aqueduct.io/docs/http/threading/). The Dart sample here uses three.

I'll put up a PR when I have a moment, or if someone can beat me to it: note that the sample put up oddly subtracts one from the number of isolates in the for loop, so just updating that isolate count variable isn't accurate.