Re: [naviserver-devel] keepalive performance issue

[Gustaf N. <ne...@wu...>]

On 30.11.18 21:17, Jeff Rogers wrote:
> Ok, thickening the plot a little bit - if I enable adp parsing and 
> serve the exact same file as adp, the delay on localhost goes away.   
> So, something weird with plain file handling on loopback?
i tested now with debian-sid and can confirm the behavior, which
happens on linux, but not on macOS.

This is really strange:

- adp works via NsAdpPageProc(), the plain file works via Ns_FastPathProc()
    but the driver calls are the same.

- Ns_FastPathProc() has actually three modes, "mmap", "cache" and 
"delivery via fd"
    (last is default). The behavior is the same in all three modes.

- when configuring writerthreads (which is recommended), and one 
requests content larger than
    1K then the behavior disappears.

    The 1k lower limit is currently hard-coded, under the assumption for 
very small
    chunks, the context switch overhead might be more expensive than a 
potential
    gain of the writer thread.

- The behavior lust like this on loopback devices.

There is no special handling of loopback devices in NaviServer.

-gn