netdev - Re: [PATCH net-next] net: skbuff: Skip early return in skb

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <9b084601-9d64-4737-8c32-4c295aafd3df@redhat.com>
Date: Thu, 1 Aug 2024 11:07:07 +0200
From: Paolo Abeni <pabeni@...hat.com>
To: Breno Leitao <leitao@...ian.org>
Cc: "David S. Miller" <davem@...emloft.net>,
 Eric Dumazet <edumazet@...gle.com>, Jakub Kicinski <kuba@...nel.org>,
 leit@...a.com, Chris Mason <clm@...com>,
 "open list:NETWORKING DRIVERS" <netdev@...r.kernel.org>,
 open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH net-next] net: skbuff: Skip early return in skb_unref when
 debugging

On 7/31/24 13:24, Breno Leitao wrote:
> Hello Paolo,
> 
> On Tue, Jul 30, 2024 at 11:38:38AM +0200, Paolo Abeni wrote:
>> Could you please benchmark such scenario before and after this patch?
> 
> I've tested it on a 18-core Xeon D-2191A host, and I haven't found any
> different in either TX/RX in TCP or UDP. At the same time, I must admit
> that I have very low confidence in my tests.
> 
> I run the following tests for 10x on the same machine, just changing my
> patch, and I getting the simple average of these 10 iterations. This is
> what I am doing for TCP and UDP:
> 
> TCP:
> 	# iperf -s &
> 	# iperf -u -c localhost
> 
> 	Output: 16.5 Gbits/sec
> 
> UDP:
> 	# iperf -s -u &
> 	# iperf -u -c localhost
> 
> 	Output: 1.05 Mbits/sec
> 
> I don't know how to explain why UDP numbers are so low. I am happy to
> run different tests, if you have any other recommendation.

Beyond the '-b 0' argument, as noted by Jason, you need to do manual CPU 
pinning of both the sender and the receiver. Additionally, to really 
flood the receiver you likely have to place the sender on a different host.

In any case, given all the prior discussion, I don't intend to block 
this patch.

Cheers,

Paolo