linux-kernel - Re: [PATCH v24 00/14] Subject: Introduce Data Access MONitor (DAMON)

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]

Date:   Wed, 24 Feb 2021 14:30:05 +0100
From:   SeongJae Park <sjpark@...zon.com>
To:     SeongJae Park <sjpark@...zon.com>
CC:     <akpm@...ux-foundation.org>, <Jonathan.Cameron@...wei.com>,
        <aarcange@...hat.com>, <acme@...nel.org>,
        <alexander.shishkin@...ux.intel.com>, <amit@...nel.org>,
        <benh@...nel.crashing.org>, <brendan.d.gregg@...il.com>,
        <brendanhiggins@...gle.com>, <cai@....pw>,
        <colin.king@...onical.com>, <corbet@....net>, <david@...hat.com>,
        <dwmw@...zon.com>, <elver@...gle.com>, <fan.du@...el.com>,
        <foersleo@...zon.de>, <gthelen@...gle.com>, <irogers@...gle.com>,
        <jolsa@...hat.com>, <kirill@...temov.name>, <mark.rutland@....com>,
        <mgorman@...e.de>, <minchan@...nel.org>, <mingo@...hat.com>,
        <namhyung@...nel.org>, <peterz@...radead.org>,
        <rdunlap@...radead.org>, <riel@...riel.com>, <rientjes@...gle.com>,
        <rostedt@...dmis.org>, <rppt@...nel.org>, <sblbir@...zon.com>,
        <shakeelb@...gle.com>, <shuah@...nel.org>, <sj38.park@...il.com>,
        <snu@...zon.de>, <vbabka@...e.cz>, <vdavydov.dev@...il.com>,
        <yang.shi@...ux.alibaba.com>, <ying.huang@...el.com>,
        <zgf574564920@...il.com>, <linux-damon@...zon.com>,
        <linux-mm@...ck.org>, <linux-doc@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v24 00/14] Subject: Introduce Data Access MONitor (DAMON)

On Thu, 4 Feb 2021 16:31:36 +0100 SeongJae Park <sjpark@...zon.com> wrote:

> From: SeongJae Park <sjpark@...zon.de>
> 
[...]
> 
> Introduction
> ============
> 
> DAMON is a data access monitoring framework for the Linux kernel.  The core
> mechanisms of DAMON called 'region based sampling' and 'adaptive regions
> adjustment' (refer to 'mechanisms.rst' in the 11th patch of this patchset for
> the detail) make it
> 
>  - accurate (The monitored information is useful for DRAM level memory
>    management. It might not appropriate for Cache-level accuracy, though.),
>  - light-weight (The monitoring overhead is low enough to be applied online
>    while making no impact on the performance of the target workloads.), and
>  - scalable (the upper-bound of the instrumentation overhead is controllable
>    regardless of the size of target workloads.).
> 
> Using this framework, therefore, several memory management mechanisms such as
> reclamation and THP can be optimized to aware real data access patterns.
> Experimental access pattern aware memory management optimization works that
> incurring high instrumentation overhead will be able to have another try.
> 
> Though DAMON is for kernel subsystems, it can be easily exposed to the user
> space by writing a DAMON-wrapper kernel subsystem.  Then, user space users who
> have some special workloads will be able to write personalized tools or
> applications for deeper understanding and specialized optimizations of their
> systems.
> 

I realized I didn't introduce a good, intuitive example use case of DAMON for
profiling so far, though DAMON is not for only profiling.  One straightforward
and realistic usage of DAMON as a profiling tool would be recording the
monitoring results with callstack and visualize those by timeline together.

For example, below link shows that visualization for a realistic workload,
namely 'fft' in SPLASH-2X benchmark suite.  From that, you can know there are
three memory access bursting phases in the workload and
'FFT1DOnce.cons::prop.2()' looks responsible for the first and second hot
phase, while 'Transpose()' is responsible for the last one.  Now the programmer
can take a deep look in the functions and optimize the code (e.g., adding
madvise() or mlock() calls).

    https://damonitor.github.io/temporal/damon_callstack.png

We used the approach for 'mlock()'-based optimization of a range of other
realistic benchmark workloads.  The optimized versions achieved up to about
2.5x performance improvement under memory pressure[1].

Note: I made the uppermost two figures in above 'fft' visualization (working
set size and access frequency of each memory region by time) via the DAMON user
space tool[2], while the lowermost one (callstack by time) is made using perf
and speedscope[3].  We have no descent and totally automated tool for that yet
(will be implemented soon, maybe under perf as a perf-script[4]), but you could
reproduce that with below commands.

    $ # run the workload
    $ sudo damo record $(pidof <your_workload>) &
    $ sudo perf record -g $(pidof <your_workload>)
    $ # after your workload finished (you should also finish perf on your own)
    $ damo report wss --sortby time --plot wss.pdf
    $ damo report heats --heatmap freq.pdf
    $ sudo perf script | speedscope -
    $ # open wss.pdf and freq.pdf with our favorite pdf viewer

[1] https://linuxplumbersconf.org/event/4/contributions/548/attachments/311/590/damon_ksummit19.pdf
[2] https://lore.kernel.org/linux-mm/20201215115448.25633-8-sjpark@amazon.com/
[3] https://www.speedscope.app/
[4] https://lore.kernel.org/linux-mm/20210107120729.22328-1-sjpark@amazon.com/