netdev - Re: Performance comparision / qdisc / no qdisc / queueing linux kernel 4.12

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

Date:   Sat, 12 Aug 2017 09:23:24 -0700
From:   Tom Herbert <tom@...bertland.com>
To:     Stephen Hemminger <stephen@...workplumber.org>
Cc:     Paweł Staszewski <pstaszewski@...are.pl>,
        Linux Kernel Network Developers <netdev@...r.kernel.org>
Subject: Re: Performance comparision / qdisc / no qdisc / queueing linux
 kernel 4.12

On Fri, Aug 11, 2017 at 4:16 PM, Stephen Hemminger
<stephen@...workplumber.org> wrote:
> Any of the current qdisc that do rate control are single threaded.
> Since bandwidth is a shared resource, the qdisc has to acquire a lock
> and update cache shared data structures.
>
> Don't expect HTB or HFSC to be free.
>
This is a well known problem and there have been several attempts to
address it over the years (a long time ago at Google I implemented a
horrible hack called "hold queue" that put a parallelized queue in
front of HTB. This eliminated the lock contention, but not without
side effects).

Now there's a new scheduler proposed by some Googlers called Carousel
that promises scalable host traffic shaping
(https://www.cc.gatech.edu/~amsmti3/files/carousel-sigcomm17.pdf).
Looks like they didn't implement a kernel Qdisc, but might be
interesting for someone to look into...

Tom

>>
>> 4.13-rc3 have kernel panic when deleting qdisc - will try to get panic trace
>
> Please do.
>