netdev
[Top] [All Lists]

Re: [Fwd: Re: possible bug x86 2.4.2 SMP in IP receive stack]

To: kuznet@xxxxxxxxxxxxx
Subject: Re: [Fwd: Re: possible bug x86 2.4.2 SMP in IP receive stack]
From: Andi Kleen <ak@xxxxxx>
Date: Thu, 8 Mar 2001 14:41:34 +0100
Cc: Bob Felderman <feldy@xxxxxxxx>, andrewm@xxxxxxxxxx, davem@xxxxxxxxxx, netdev@xxxxxxxxxxx, pp@xxxxxxxxxxxxxx
In-reply-to: <200103071848.VAA11377@xxxxxxxxxxxxx>; from kuznet@xxxxxxxxxxxxx on Wed, Mar 07, 2001 at 07:48:25PM +0100
References: <200103071834.KAA13649@xxxxxxxx> <200103071848.VAA11377@xxxxxxxxxxxxx>
Sender: owner-netdev@xxxxxxxxxxx
On Wed, Mar 07, 2001 at 07:48:25PM +0100, A.N.Kuznetsov wrote:
> Hello!
> 
> > is down quite a bit.
> 
> Actually, the fact that eviction happens during your tests implies
> that you should increase eviction threshold (ipfrag_high_thresh)
> to value, when eviction never happens on LAN tests. It should happen
> only if driver has lost some fragment... not good on LAN too.

I think the current default size of the defrag buffer is far too low anyways.
It should probably scale by numbers of interfaces also and better by 
bandwidth of the interface. 

Something like 1-2MB is much more reasonable with today's memory sizes.

In my experience upping it can increase NFS performance a lot.


-Andi
-- 
This is like TV. I don't like TV.

<Prev in Thread] Current Thread [Next in Thread>