From: Christian Stroetmann <stroetmann@ontolab.com>
To: Daniel Phillips <daniel@phunq.net>
Cc: Howard Chu <hyc@symas.com>,
Mike Galbraith <umgwanakikbuti@gmail.com>,
Dave Chinner <david@fromorbit.com>,
linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
tux3@tux3.org, Theodore Ts'o <tytso@mit.edu>,
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Subject: Re: xfs: does mkfs.xfs require fancy switches to get decent performance? (was Tux3 Report: How fast can we fsync?)
Date: Thu, 30 Apr 2015 20:22:41 +0200 [thread overview]
Message-ID: <554272F1.80801@ontolab.com> (raw)
In-Reply-To: <554246D7.40105@phunq.net>
On the 30th of April 2015 17:14, Daniel Phillips wrote:
Hallo hardcore coders
> On 04/30/2015 07:28 AM, Howard Chu wrote:
>> Daniel Phillips wrote:
>>>
>>> On 04/30/2015 06:48 AM, Mike Galbraith wrote:
>>>> On Thu, 2015-04-30 at 05:58 -0700, Daniel Phillips wrote:
>>>>> On Thursday, April 30, 2015 5:07:21 AM PDT, Mike Galbraith wrote:
>>>>>> On Thu, 2015-04-30 at 04:14 -0700, Daniel Phillips wrote:
>>>>>>
>>>>>>> Lovely sounding argument, but it is wrong because Tux3 still beats XFS
>>>>>>> even with seek time factored out of the equation.
>>>>>> Hm. Do you have big-storage comparison numbers to back that? I'm no
>>>>>> storage guy (waiting for holographic crystal arrays to obsolete all this
>>>>>> crap;), but Dave's big-storage guy words made sense to me.
>>>>> This has nothing to do with big storage. The proposition was that seek
>>>>> time is the reason for Tux3's fsync performance. That claim was easily
>>>>> falsified by removing the seek time.
>>>>>
>>>>> Dave's big storage words are there to draw attention away from the fact
>>>>> that XFS ran the Git tests four times slower than Tux3 and three times
>>>>> slower than Ext4. Whatever the big storage excuse is for that, the fact
>>>>> is, XFS obviously sucks at little storage.
>>>> If you allocate spanning the disk from start of life, you're going to
>>>> eat seeks that others don't until later. That seemed rather obvious and
>>>> straight forward.
>>> It is a logical falacy. It mixes a grain of truth (spreading all over the
>>> disk causes extra seeks) with an obvious falsehood (it is not necessarily
>>> the only possible way to avoid long term fragmentation).
>> You're reading into it what isn't there. Spreading over the disk isn't (just) about avoiding
>> fragmentation - it's about delivering consistent and predictable latency. It is undeniable that if
>> you start by only allocating from the fastest portion of the platter, you are going to see
>> performance slow down over time. If you start by spreading allocations across the entire platter,
>> you make the worst-case and average-case latency equal, which is exactly what a lot of folks are
>> looking for.
> Another fallacy: intentionally running slower than necessary is not necessarily
> the only way to deliver consistent and predictable latency. Not only that, but
> intentionally running slower than necessary does not necessarily guarantee
> performing better than some alternate strategy later.
>
> Anyway, let's not be silly. Everybody in the room who wants Git to run 4 times
> slower with no guarantee of any benefit in the future, please raise your hand.
>
>>>> He flat stated that xfs has passable performance on
>>>> single bit of rust, and openly explained why. I see no misdirection,
>>>> only some evidence of bad blood between you two.
>>> Raising the spectre of theoretical fragmentation issues when we have not
>>> even begun that work is a straw man and intellectually dishonest. You have
>>> to wonder why he does it. It is destructive to our community image and
>>> harmful to progress.
>> It is a fact of life that when you change one aspect of an intimately interconnected system,
>> something else will change as well. You have naive/nonexistent free space management now; when you
>> design something workable there it is going to impact everything else you've already done. It's an
>> easy bet that the impact will be negative, the only question is to what degree.
> You might lose that bet. For example, suppose we do strictly linear allocation
> each delta, and just leave nice big gaps between the deltas for future
> expansion. Clearly, we run at similar or identical speed to the current naive
> strategy until we must start filling in the gaps, and at that point our layout
> is not any worse than XFS, which started bad and stayed that way.
>
> Now here is where you lose the bet: we already know that linear allocation
> with wrap ends horribly right? However, as above, we start linear, without
> compromise, but because of the gaps we leave, we are able to switch to a
> slower strategy, but not nearly as slow as the ugly tangle we get with
> simple wrap. So impact over the lifetime of the filesystem is positive, not
> negative, and what seemed to be self evident to you turns out to be wrong.
>
> In short, we would rather deliver as much performance as possible, all the
> time. I really don't need to think about it very hard to know that is what I
> want, and what most users want.
>
> I will make you a bet in return: when we get to doing that part properly, the
> quality of the work will be just as high as everything else we have completed
> so far. Why would we suddenly get lazy?
>
> Regards,
>
> Daniel
> --
>
How?
Maybe this is explained and discussed in a new thread about allocation
or so.
Thanks
Best Regards
Have fun
C.S.
next prev parent reply other threads:[~2015-04-30 18:23 UTC|newest]
Thread overview: 160+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-04-28 23:13 Tux3 Report: How fast can we fsync? Daniel Phillips
2015-04-29 2:21 ` Mike Galbraith
2015-04-29 6:01 ` Daniel Phillips
2015-04-29 6:20 ` Richard Weinberger
2015-04-29 6:56 ` Daniel Phillips
2015-04-29 6:33 ` Mike Galbraith
2015-04-29 7:23 ` Daniel Phillips
2015-04-29 16:42 ` Mike Galbraith
2015-04-29 19:05 ` xfs: does mkfs.xfs require fancy switches to get decent performance? (was Tux3 Report: How fast can we fsync?) Mike Galbraith
2015-04-29 19:20 ` Austin S Hemmelgarn
2015-04-29 21:12 ` Daniel Phillips
2015-04-30 4:40 ` Mike Galbraith
2015-04-30 0:20 ` Dave Chinner
2015-04-30 3:35 ` Mike Galbraith
2015-04-30 9:00 ` Martin Steigerwald
2015-04-30 14:57 ` Theodore Ts'o
2015-04-30 15:59 ` Daniel Phillips
2015-04-30 17:59 ` Martin Steigerwald
2015-04-30 11:14 ` Daniel Phillips
2015-04-30 12:07 ` Mike Galbraith
2015-04-30 12:58 ` Daniel Phillips
2015-04-30 13:48 ` Mike Galbraith
2015-04-30 14:07 ` Daniel Phillips
2015-04-30 14:28 ` Howard Chu
2015-04-30 15:14 ` Daniel Phillips
2015-04-30 16:00 ` Howard Chu
2015-04-30 18:22 ` Christian Stroetmann [this message]
2015-05-11 22:12 ` Pavel Machek
2015-05-11 23:17 ` Theodore Ts'o
2015-05-12 2:34 ` Daniel Phillips
2015-05-12 5:38 ` Dave Chinner
2015-05-12 6:18 ` Daniel Phillips
2015-05-12 18:39 ` David Lang
2015-05-12 20:54 ` Daniel Phillips
2015-05-12 21:30 ` David Lang
2015-05-12 22:27 ` Daniel Phillips
2015-05-12 22:35 ` David Lang
2015-05-12 23:55 ` Theodore Ts'o
2015-05-13 1:26 ` Daniel Phillips
2015-05-13 19:09 ` Martin Steigerwald
2015-05-13 19:37 ` Daniel Phillips
2015-05-13 20:02 ` Jeremy Allison
2015-05-13 20:24 ` Daniel Phillips
2015-05-13 20:25 ` Martin Steigerwald
2015-05-13 20:38 ` Daniel Phillips
2015-05-13 21:10 ` Martin Steigerwald
2015-05-13 0:31 ` Daniel Phillips
2015-05-12 21:30 ` Christian Stroetmann
2015-05-13 7:20 ` Pavel Machek
2015-05-13 13:47 ` Elifarley Callado Coelho Cruz
2015-05-12 9:03 ` Pavel Machek
2015-05-12 11:22 ` Daniel Phillips
2015-05-12 13:26 ` Howard Chu
2015-05-11 23:53 ` Daniel Phillips
2015-05-12 0:12 ` David Lang
2015-05-12 4:36 ` Daniel Phillips
2015-05-12 17:30 ` Christian Stroetmann
2015-05-13 7:25 ` Pavel Machek
2015-05-13 11:31 ` Daniel Phillips
2015-05-13 12:41 ` Daniel Phillips
2015-05-13 13:08 ` Mike Galbraith
2015-05-13 13:15 ` Daniel Phillips
2015-04-30 14:33 ` Mike Galbraith
2015-04-30 15:24 ` Daniel Phillips
2015-04-29 20:40 ` Tux3 Report: How fast can we fsync? Daniel Phillips
2015-04-29 22:06 ` OGAWA Hirofumi
2015-04-30 3:57 ` Mike Galbraith
2015-04-30 3:50 ` Mike Galbraith
2015-04-30 10:59 ` Daniel Phillips
2015-04-30 1:46 ` Dave Chinner
2015-04-30 10:28 ` Daniel Phillips
2015-05-01 15:38 ` Dave Chinner
2015-05-01 23:20 ` Daniel Phillips
2015-05-02 1:07 ` David Lang
2015-05-02 10:26 ` Daniel Phillips
2015-05-02 16:00 ` Christian Stroetmann
2015-05-02 16:30 ` Richard Weinberger
2015-05-02 17:00 ` Christian Stroetmann
2015-05-12 17:41 ` Daniel Phillips
2015-05-12 17:46 ` Tux3 Report: How fast can we fail? Daniel Phillips
2015-05-13 22:07 ` Daniel Phillips
2015-05-26 10:03 ` Pavel Machek
2015-05-27 6:41 ` Mosis Tembo
2015-05-27 18:28 ` Daniel Phillips
2015-05-27 21:39 ` Pavel Machek
2015-05-27 22:46 ` Daniel Phillips
2015-05-28 12:55 ` Austin S Hemmelgarn
2015-05-27 7:37 ` Mosis Tembo
2015-05-27 14:04 ` Austin S Hemmelgarn
2015-05-27 15:21 ` Mosis Tembo
2015-05-27 15:37 ` Austin S Hemmelgarn
2015-05-14 7:37 ` [WIP] tux3: Optimized fsync Daniel Phillips
2015-05-14 8:26 ` [FYI] tux3: Core changes Daniel Phillips
2015-05-14 12:59 ` Rik van Riel
2015-05-15 0:06 ` Daniel Phillips
2015-05-15 3:06 ` Rik van Riel
2015-05-15 8:09 ` Mel Gorman
2015-05-15 9:54 ` Daniel Phillips
2015-05-15 11:00 ` Mel Gorman
2015-05-16 22:38 ` David Lang
2015-05-18 12:57 ` Mel Gorman
2015-05-15 9:38 ` Daniel Phillips
2015-05-27 7:41 ` Pavel Machek
2015-05-27 18:09 ` Daniel Phillips
2015-05-27 21:37 ` Pavel Machek
2015-05-27 22:33 ` Daniel Phillips
2015-05-15 8:05 ` Mel Gorman
2015-05-17 13:26 ` Boaz Harrosh
2015-05-18 2:20 ` Rik van Riel
2015-05-18 7:58 ` Boaz Harrosh
2015-05-19 4:46 ` Daniel Phillips
2015-05-21 19:43 ` [WIP][PATCH] tux3: preliminatry nospace handling Daniel Phillips
2015-05-19 14:00 ` [FYI] tux3: Core changes Jan Kara
2015-05-19 19:18 ` Daniel Phillips
2015-05-19 20:33 ` David Lang
2015-05-20 14:44 ` Jan Kara
2015-05-20 16:22 ` Daniel Phillips
2015-05-20 18:01 ` David Lang
2015-05-20 19:53 ` Rik van Riel
2015-05-20 22:51 ` Daniel Phillips
2015-05-21 3:24 ` Daniel Phillips
2015-05-21 3:51 ` David Lang
2015-05-21 19:53 ` Daniel Phillips
2015-05-26 4:25 ` Rik van Riel
2015-05-26 4:30 ` Daniel Phillips
2015-05-26 6:04 ` David Lang
2015-05-26 6:11 ` Daniel Phillips
2015-05-26 6:13 ` David Lang
2015-05-26 8:09 ` Daniel Phillips
2015-05-26 10:13 ` Pavel Machek
2015-05-26 7:09 ` Jan Kara
2015-05-26 8:08 ` Daniel Phillips
2015-05-26 9:00 ` Jan Kara
2015-05-26 20:22 ` Daniel Phillips
2015-05-26 21:36 ` Rik van Riel
2015-05-26 21:49 ` Daniel Phillips
2015-05-27 8:41 ` Jan Kara
2015-06-21 15:36 ` OGAWA Hirofumi
2015-06-23 16:12 ` Jan Kara
2015-07-05 12:54 ` OGAWA Hirofumi
2015-07-09 16:05 ` Jan Kara
2015-07-31 4:44 ` OGAWA Hirofumi
2015-07-31 15:37 ` Raymond Jennings
2015-07-31 17:27 ` Daniel Phillips
2015-07-31 18:29 ` David Lang
2015-07-31 18:43 ` Daniel Phillips
2015-07-31 22:12 ` Daniel Phillips
2015-07-31 22:27 ` David Lang
2015-08-01 0:00 ` Daniel Phillips
2015-08-01 0:16 ` Daniel Phillips
2015-08-03 13:07 ` Jan Kara
2015-08-01 10:55 ` Elifarley Callado Coelho Cruz
2015-08-18 16:39 ` Rik van Riel
2015-08-03 13:42 ` Jan Kara
2015-08-09 13:42 ` OGAWA Hirofumi
2015-08-10 12:45 ` Jan Kara
2015-08-16 19:42 ` OGAWA Hirofumi
2015-05-26 10:22 ` Sergey Senozhatsky
2015-05-26 12:33 ` Jan Kara
2015-05-26 19:18 ` Daniel Phillips
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=554272F1.80801@ontolab.com \
--to=stroetmann@ontolab.com \
--cc=daniel@phunq.net \
--cc=david@fromorbit.com \
--cc=hirofumi@mail.parknet.co.jp \
--cc=hyc@symas.com \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=tux3@tux3.org \
--cc=tytso@mit.edu \
--cc=umgwanakikbuti@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).