HTTP2 Multiplexing

Hi

Different from HTTP1.1 pipeline, HTTP2 allows multiple request and
response
messages to be in flight at the same time. I was wondering what the
strategy Nginx adopt to implement this main feature.

Is every single stream correspond to a thread. If not, how can Nginx
provide multiple parallel requests handling. If you can locate the
correspond code for me, that would be great

Best Regards
Muhui Jiang

On Wednesday 04 May 2016 11:25:11 Muhui Jiang wrote:

Hi

Different from HTTP1.1 pipeline, HTTP2 allows multiple request and response
messages to be in flight at the same time. I was wondering what the
strategy Nginx adopt to implement this main feature.

Nginx allows multiple request and responses in multiple connections
using
HTTP/1.x as well. HTTP/2 changes nothing here (except it uses only one
connection, but it’s not important from the basic architecture point of
view).

Is every single stream correspond to a thread. If not, how can Nginx
provide multiple parallel requests handling. If you can locate the
correspond code for me, that would be great

No, nginx uses asynchronous non-blocking event-driven architecture
instead of mapping requests into separate threads.

It have used multiplexing of requests handling in single process many
years before HTTP/2 was invented.

A more detailed explanation can be found here:

wbr, Valentin V. Bartenev

Hi

Nginx allows multiple request and responses in multiple connections using
HTTP/1.x as well. HTTP/2 changes nothing here (except it uses only one
connection, but it’s not important from the basic architecture point of
view).

If so, it seems there is no difference or improvement of the
implementation
on the feature of multiplexing compared with Http/1.1 pipeline.

How do you solve the problem Head-of-line blocking occurred in Http/1.1.

Best Regards
Muhui Jiang

2016-05-04 18:19 GMT+08:00 Valentin V. Bartenev [email protected]:

On Wednesday 04 May 2016 18:50:44 Muhui Jiang wrote:

How do you solve the problem Head-of-line blocking occurred in Http/1.1.

[…]

You have mixed multiplexing of data transferring over single connection,
which is the feature of HTTP/2, with multiplexing of requests processing
inside one process, which is the feature of nginx.

These things have no relation to each other.

Please, read the article I’ve pointed out, nginx doesn’t need separate
threads to process something in parallel.

The Head-of-line blocking problem in HTTP/1.1 is presented only if you
use limited number of connections, and HTTP/2 solves it in some degree
but puts the limit and the problem on different level.

With HTTP/2 now you have limit of virtual streams inside HTTP/2
connection and the HOL problem on TCP level.

wbr, Valentin V. Bartenev

Hi

Thanks Valentin. I’ve know the principle

Best Regards
Muhui Jiang

2016-05-04 19:18 GMT+08:00 Valentin V. Bartenev [email protected]: