lkml.org 
[lkml]   [1996]   [Mar]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectPPP slowdown (Was Re: No Interrupt latency but too many interrupts)
On 27 Mar 1996, Christoph Lameter wrote:

> I am looking since weeks for the source of the stuttering PPP and SLIP
> performance and I guess I found something.

I'm looking into this too. I've started my investigations at kernel
1.3.60, and have checked all the way through to 1.3.79 by looking at the
patches - I don't have time to compile them all and test every single one!

> I observed that my system still has a load of 0.30 when totally idle
> so I had a look at the interrupts and although the serial port
> was not in use there were still a couple of hundred interrupts
> per second.

I think this may be a separate problem. My loadav is usually 0.01 to 0.03
but I still get PPP slowdown.

> How could the serial port generate so many interrupts while not in use?

Where do you get this info ? /proc/interrupts doesn't seem to report this.

> Were there any modifications to the serial driver lately?

In my searches through the kernel patches I mentioned, I have found only
two major changes within the serial/ppp code. These are the addition of
IPX (which I think is the likely culprit), and the addition of some
additional ioctls. All of this is part of the 1.3.72 patch which is about
when the complaints started (correct me if that's wrong).

The rest of the hundreds of lines of patches to the ppp subsystem has no
effect worth mentioning and is made up of:
cleanups
typedef/#define moves
comments changes
whitespace/tab cleanups
changes in initialisation/shutdown code

I didn't notice any major changes to the serial driver code although a
couple of race conditions were removed. I could be very wrong about this
though.

I might have missed somethig of significance of-course, and I'll keep
looking.

All comments welcome.
Steve



\
 
 \ /
  Last update: 2005-03-22 13:36    [W:0.088 / U:1.788 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site