Subject: Re: high input packet rate can lead to process starvation
To: YAMAMOTO Takashi <>
From: Tad Hunt <>
List: tech-net
Date: 09/29/2006 13:07:07
If it were just a simple matter of importing YAMAMOTO Takashi's changes, 
and bringing them up-to-date with NetBSD-current, I would do so.

However in our application of NetBSD, the performance of the networking 
stack is pretty much irrelevant, so most any solution that allows user 
processes to run under high packet loads is acceptable.

Certainly this is not true for many applications.

Unfortunately, we do not have the resources (time for implementation and 
test or hardware) available to do extensive analysis to determine what 
kind of performance impact these changes would have on the stack.


YAMAMOTO Takashi wrote:
>>>> What are the chances of something like this making it into NetBSD  
>>>> in the
>>>> near future?
>>>> -Tad
>>> i believe something is necessary here.
>>> so if someone (eg. you :-) completed it in a reasonable way,
>>> it's likely made into the tree, i think.  (at least i'll support it.)
>>> i don't think i have enough time to finish my version in the near  
>>> future.
>> What's missing from your version?
> - test and investigation.  although it worked in some extent for my load
>   when i wrote it, i'm not sure if it's a good enough way to deal with
>   the problem.
> - port to -current.
> - i vaguely remember that there are kmem grovellers around the queues.
> YAMAMOT Takashi