Httpserver performance

[Rob Bygrave]

> out of curiosity

I hope it becomes more than that :) For example, I'd say that today a
HttpServer does not need to support the Servlet Spec (e.g. Helidon 4 SE
HttpServer), and all that I'd look to add on top of JDK HttpServer is a
"Routing Layer".  A lot of deployed services are not directly
internet facing but instead have gateways in front of them so imo we
frequently don't need more http features (like http3 etc).

Getting these changes in and JDK HttpServer could be pretty competitive for
a lot of cases due to its minimalism.


On Thu, 11 Jul 2024 at 23:34, robert engels <robaho at me.com> wrote:

> It was done more out of curiosity. I know the code fairly well so I was
> surprised in the low performance of the JDK httpserver so I set about to
> figure out why.
>
> > On Jul 11, 2024, at 4:17 AM, Alan Bateman <Alan.Bateman at oracle.com>
> wrote:
> >
> > On 05/07/2024 18:41, Robert Engels wrote:
> >> Some that follow this list might be interested in the latest release of
> my httpserver - which is a port of the JDK one - designed for virtual
> threads (no async, no synchronized blocks) at
> https://github.com/robaho/httpserver
> >>
> >> It shows nearly 3x performance on the tech empower low-level benchmarks.
> >>
> >> I attribute most of the performance gain due to it supporting http
> pipelining, which I think should be easy to backport in the JDK version.
> There are some other micro improvements that help quite a bit for these
> types of tests. As it is, it is able to saturate a 1gbit ethernet link.
> >
> > Thanks for the post, it's interesting to this fork being used for
> performance testing. As you probably know, the original motive for this
> HTTP server was something basic to support XML web services callbacks.
> Today it provides an on-ramp to web development with local testing. It's
> also useful for other testing too, or just to serve up API docs or other
> content. It's never been a goal to have it be competitive in either
> features or performances when compared to other HTTP servers. Just saying
> to explain why there hasn't been a interest on net-dev to add pipelining or
> other features.
> >
> > -Alan
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/loom-dev/attachments/20240712/7f025095/attachment.htm>