[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5ef10a85-3b2a-468e-8a67-200c6ad63dfe@redhat.com>
Date: Mon, 1 Sep 2025 09:59:15 +0200
From: Paolo Abeni <pabeni@...hat.com>
To: John Ousterhout <ouster@...stanford.edu>
Cc: "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
Eric Dumazet <edumazet@...gle.com>, Simon Horman <horms@...nel.org>,
Jakub Kicinski <kuba@...nel.org>
Subject: Re: [PATCH net-next v15 03/15] net: homa: create shared Homa header
files
On 8/29/25 7:08 PM, John Ousterhout wrote:
> On Fri, Aug 29, 2025 at 12:53 AM Paolo Abeni <pabeni@...hat.com> wrote:
>> If that is percent of total CPU time for a single core, such value is
>> inconsistent with my benchmarking where a couple of timestamp() reads
>> per aggregate packet are well below noise level.
>
> Homa is doing a lot more than a couple of timestamp() reads per
> aggregate packet.
Than it looks like this is the problem. Data processing should require ~
a single ts per packet. If you need more for instrumentation, you should
likely put such code behind a compiler's conditional and enable them
just in devel/debug build.
Or even better you could use ftrace/bpf trace for that.
/P
Powered by blists - more mailing lists