netdev - Heavy tcp_send_ack in recvfrom syscall

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>] [day] [month] [year] [list]

Message-ID: <CAD_cgaze+bbuMsY8BRuYksJEGKWjt2RsJVBhV_psMWxb84Fzaw@mail.gmail.com>
Date: Thu, 6 Mar 2025 01:20:28 +0300
From: Sergey Melnikov <melnikov.sergey.v@...il.com>
To: netdev@...r.kernel.org
Subject: Heavy tcp_send_ack in recvfrom syscall

Hello,

I'm digging one issue(not sure if it's an issue and not sure how to
fix this) with linux tcp stack. Our application receives a big number
of small packets. Total network utilization isn't really big (dozens
of Mbit/s). Platform: ubuntu 24, ENA-enabled AWS VM.

I observe tcp_send_ack takes half of the whole 'recvfrom' cpu time
(with callstacks to AWS ENA driver). Total recvfrom syscall takes
approx 8.8% and tcp_send_ack is about 5% of the whole profile.

I tried to extend the sockets buffer with proper sysctl's
(net.ipv4.tcp_wmem/net.ipv4.tcp_rmem) and got some improvements.
Nevertheless, I still can't understand how to send ACKs less
frequently and reduce CPU consumption and latency here.

I've found the TCP_QUICKACK socket flag. According to man pages,
disabling this flag will do exactly I need. Will it fix my issue with
manu ACKs?

Is there any ideas or advice on what to do and how to fix such an issue?

--Sergey