Hi Jeanfrancois.
The configuration dumped out by Grizzly during these tests was the following from grizzly-http-webserver-1.9.3.jar:
[java] Grizzly configuration for port 9090
[java] Thread Pool: StatsThreadPool[name=http, priority=5, min-threads=5, max-threads=5, max-queue-size=2147483647, initial-byte-buffer-size=8192, byte-buffer-type=HEAP_VIEW, is-shutdown=false, port=9090]
[java] ByteBuffer size: 8192
[java] maxHttpHeaderSize: 8192
[java] maxKeepAliveRequests: 256
[java] keepAliveTimeoutInSeconds: 30
[java] Static File Cache enabled: true
[java] Static resources directory: /home/niall/Workspace/security/work/Grizzly/1.9.3
[java] Adapter : com.sun.grizzly.tcp.StaticResourcesAdapter
[java] Asynchronous Request Processing enabled: false
I realize this may not be optimal, if you can provide a better config or a template from which to run Grizzly that would be great.
Also I know Simple and Grizzly can be configured for higher performance, for instance Simple can have the write segment sizes for asynchronous writes configured, as well as threading strategies. I know Grizzly has similar config options.
I assure you my tests will be objective, for good measure I am going to check in the Faban test harness to the Simple subversion repository for interrogation. I will include the configs and code for both servers.
Lets agree on 3 or 4 real world scenarios underpinned by a single server configuration for each server to demonstrate how they both behave. For example something akin to what Jersey or Restlet would produce would be good e.g dynamic content. Also, for good measure I will include Netty as you have mentioned.
I hope you are happy with this criteria. I will also wait for any response Grizzly may have to the tests before I publish the tests results.
Thanks,
Niall
--- On Mon, 2/2/09, Niall Gallagher <gallagher_niall_at_yahoo.com> wrote:
> From: Niall Gallagher <gallagher_niall_at_yahoo.com>
> Subject: Re: Grizzly Strategies
> To: dev_at_grizzly.dev.java.net
> Date: Monday, February 2, 2009, 12:24 PM
> Hi,
>
> Faban results are slightly different, however it favours
> Simple, which is almost twice as fast as Grizzly 1.9.3. I
> have used the same config, with no JVM args to favour either
> server. Here is what I see.
>
> Simple:
> <metric
> unit="ops/sec">7855.375</metric>
>
> Grizzly:
> <metric
> unit="ops/sec">4061.683</metric>
>
> Can you provide a good template for starting Grizzly with
> its best possible config. I have attached the Faban
> summaries.
>
> Thanks,
> Niall
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail:
> dev-unsubscribe_at_grizzly.dev.java.net
> For additional commands, e-mail:
> dev-help_at_grizzly.dev.java.net