Re: [boost] [beast] Request for Discussion

Subject: Re: [boost] [beast] Request for Discussion
From: Vinnie Falco (vinnie.falco_at_[hidden])
Date: 2017-06-10 02:28:10

Reanimating an old topic about Beast HTTP

On Wed, Sep 28, 2016 at 5:39 PM, David Sankel <camior_at_[hidden]> wrote:
> I don't think a synchronous interface is necessary.

I disagree. Synchronous interface used with the non-blocking setting,
and a reactor style wait operation
(http://www.boost.org/doc/libs/1_64_0/doc/html/boost_asio/overview/core/reactor.html)
present a strong alternative to the traditional proactor model.

Specifically, using the reactor style wait operation means that no
memory buffers need to be allocated while a connection is idle /
waiting for bytes to become available. A robust reactor style
implementation can service tens of thousands of connections using
memory proportional to the number of threads rather than the number of
connections.

The technique is overlooked because of the assumption that the
asynchronous interfaces should be preferred, and the synchronous
interfaces deprecated. This impressive performance feat is not
theoretical, it is already happening in the wild, for example, this
library uses the reactor model and gets impressive performance
figures:
https://github.com/uNetworking/uWebSockets

I will be investigating these techniques in Beast, most likely as an
example server for resource-constrained devices, you can follow the
progress here:
https://github.com/vinniefalco/Beast/issues/445