netdev - Re: [PATCH net-next v4 0/3] add tx packets aggregation to ethtool and rmnet

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: <20230113114800.357a96e2@kernel.org>
Date:   Fri, 13 Jan 2023 11:48:00 -0800
From:   Jakub Kicinski <kuba@...nel.org>
To:     Daniele Palmas <dnlplm@...il.com>
Cc:     Alexander Duyck <alexander.duyck@...il.com>,
        David Miller <davem@...emloft.net>,
        Paolo Abeni <pabeni@...hat.com>,
        Eric Dumazet <edumazet@...gle.com>,
        Subash Abhinov Kasiviswanathan <quic_subashab@...cinc.com>,
        Sean Tranchetti <quic_stranche@...cinc.com>,
        Jonathan Corbet <corbet@....net>,
        Alexander Lobakin <alexandr.lobakin@...el.com>,
        Gal Pressman <gal@...dia.com>, Dave Taht <dave.taht@...il.com>,
        Bjørn Mork <bjorn@...k.no>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        netdev@...r.kernel.org
Subject: Re: [PATCH net-next v4 0/3] add tx packets aggregation to ethtool
 and rmnet

On Fri, 13 Jan 2023 08:16:48 -0800 Alexander Duyck wrote:
> > ETHTOOL_A_COALESCE_TX_AGGR_MAX_FRAMES works also as a way to determine
> > that tx aggregation has been enabled by the userspace tool managing
> > the qmi requests, otherwise no aggregation should be performed.  
> 
> Is there a specific reason why you wouldn't want to take advantage of
> aggregation that is already provided by the stack in the form of
> things such as GSO and the qdisc layer? I know most of the high speed
> NICs are always making use of xmit_more since things like GSO can take
> advantage of it to increase the throughput. Enabling BQL is a way of
> taking that one step further and enabling the non-GSO cases.

The patches had been applied last night by DaveM but FWIW I think
Alex's idea is quite interesting. Even without BQL I believe you'd 
get xmit_more set within a single GSO super-frame. TCP sends data
in chunks of 64kB, and you're only aggregating 32kB. If we could
get the same effect without any added latency and user configuration
that'd be great.