Serving big files at constant rates

Re: Serving big files at constant rates

itpp2012 — Tue, 15 May 2012 06:37:03 -0400

Could very well be a lan driver issue, rx/tx values too high or rx value higher then tx, flow control not used or used but not supported, etc..... try an older driver or if its really old a newer one. Also see if you can get a driver with debugging compiled in or support for raw mode and run something like wireshark to see whats going on.

Re: Serving big files at constant rates

B.R. — Mon, 14 May 2012 23:10:00 -0400

Hi,

I tried that, but I only see these consequences:
- Higher peek transfer speed (up to 1.1/1.2 MiB/s)
- Higher instability (bumps down to 300/400 KiB/s and with longer duration)
- Lower average transfer rate overall

Well, does that mean I have some network interfaces trouble?
---
*B. R.*

On Mon, May 14, 2012 at 12:29 PM, itpp2012 wrote:

> Why not create a 1gb ramdrive to make sure it is a disk IO issue. It
> might be a lan driver buffer issue.
>
> Posted at Nginx Forum:
> http://forum.nginx.org/read.php?2,226392,226422#msg-226422
>
> _______________________________________________
> nginx mailing list
> nginx@nginx.org
> http://mailman.nginx.org/mailman/listinfo/nginx
>
_______________________________________________
nginx mailing list
nginx@nginx.org
http://mailman.nginx.org/mailman/listinfo/nginx

Re: Serving big files at constant rates

itpp2012 — Mon, 14 May 2012 12:29:33 -0400

Why not create a 1gb ramdrive to make sure it is a disk IO issue. It might be a lan driver buffer issue.

Serving big files at constant rates

B.R. — Mon, 14 May 2012 09:46:01 -0400

Hi,

A recent thread kept my attention recently:
http://forum.nginx.org/read.php?2,226327

I don't have such a big trouble, but I don't see how to solve it.

My server usually serves small files suchas classical web pages.
On a particular location, though, I serve big files (around 100MB+).

I noticed that when those files were served, there was 'bumps' in the
transfer rate, whether the trasnfer stalled for 1s or the transfer rate
downgraded to 500Kb/s during the same period.
Those bumps were monitored server-side, so it is not some trouble in any
network in the middle.
I ensured that the only service using significant bandwith would be Nginx.

I guess I am experiencing some disk I/O speed limitations.
Thus, I tried to use aio on that location:
aio on;
directio 4m;
output_buffers 1m; (I tried 512k first hand)

The bumps are still there, appearing randomly but quite often and regularly
on the long shot.

What kind of I/O trouble is that?

Thanks,
---
*B. R.*
_______________________________________________
nginx mailing list
nginx@nginx.org
http://mailman.nginx.org/mailman/listinfo/nginx