linux-kernel - [Question]: Tail latency problem in scheduler

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: <CAKiGEcL6=RgbA6__zU54hSJo4kG8LWi61ETtWozTUwKpZFBdmQ@mail.gmail.com>
Date:   Tue, 3 Oct 2017 18:44:46 +0330
From:   Afshin Jamshidi <jamshidi.afshin@...il.com>
To:     linux-kernel@...r.kernel.org
Subject: [Question]: Tail latency problem in scheduler

According to https://research.google.com/pubs/pub40801.html article,
tail delay is problematic in enterprise service providers.
According to https://dl.acm.org/citation.cfm?id=2670988 article, FIFO
scheduling works best and LIFO is worst for tail latency.
As we test Linux scheduler called CFS, behaves mostly LIFO. Small
changes to CFS makes it FIFO style and tail delay reduces to 50%.
I want to argue about CFS strategy. It seems no body cares about tail
latency at kernel level.
How enterprise service providers overcome tail latency problem?
Is it possible to at least have a flag that changes scheduler strategy
for I/O intensive multi threaded applications?