The Unix Heritage Society mailing list
 help / color / mirror / Atom feed
From: "Theodore Ts'o" <tytso@mit.edu>
To: Jon Steinhart <jon@fourwinds.com>
Cc: The Unix Heretics Society mailing list <tuhs@minnie.tuhs.org>
Subject: Re: [TUHS] Is it time to resurrect the original dsw (delete with switches)?
Date: Thu, 2 Sep 2021 12:57:50 -0400	[thread overview]
Message-ID: <YTECjlhuKlC4aIad@mit.edu> (raw)
In-Reply-To: <202109021552.182FqNLB3750785@darkstar.fourwinds.com>

On Thu, Sep 02, 2021 at 08:52:23AM -0700, Jon Steinhart wrote:
> Long intro, on to the question.  Anyone know what it does to reliability to
> spin disks up and down.  I don't really need the media disks to be constantly
> spinning; when whatever I'm listening to in the evening finishes the disk
> could spin down until morning to save energy.  Likewise the video disk drive
> is at most used for a few hours a day.
> 
> My big disks (currently 16T and 12T) bake when they're spinning which can't
> be great for them, but I don't know how that compares to mechanical stress
> from spinning up and down from a reliability standpoint.  Anyone know?

First of all, I wouldn't worry too much about big disks "baking" while
they are spinning.  Google runs its disks hot, in data centers where
the ambient air temperatures is at least 80 degrees Fahrenheit[1], and
it's not alone; Dell said in 2012 that it would honor warranties for
servers running in environments as hot as 115 degrees Fahrenheit[2].

[1] https://www.google.com/about/datacenters/efficiency/
[2] https://www.datacenterknowledge.com/archives/2012/03/23/too-hot-for-humans-but-google-servers-keep-humming

And of course, if the ambient *air* temperature is 80 degrees plus,
you can just imagine the temperature at the hard drive.

It's also true that a long time ago, disk drives had limited number of
spin up/down cycles; this was in its spec sheet, and SMART would track
the number of disk spinups.  I had a laptop drive where I had
configured the OS so it would spin down the drive after 30 seconds of
idle, and I realized that after about 9 months, SMART stats had
reported that I had used up almost 50% of the rated spin up/down
cycles for a laptop drive.  Needless to say, I backed of my agressive
spindown policies.

That being said, more modern HDD's have been designed for better power
effiencies, with slower disk rotational speeds (which is probably fine
for media disks, unless you are serving a large number of different
video streams at the same time), and they are designed to allow for a
much larger number of spindown cycles.  Check your spec sheets, this
will be listed as load/unload cycles, and it will typically be a
number like 200,000, 300,000 or 600,000.

If you're only spinning down the a few times a day, I suspect you'll
be fine.  Especially since if the disk dies due to a head crash or
other head failure, it'll be a case of an obvious disk failure, not
silent data corruption, and you can just pull your backups out of your
fire safe.


I don't personally have a lot of knowledge of how modern HDD's
actually survive large numbers of load/unload cycles, because at $WORK
we keep the disks spinning at all times.  A disk provides value in two
ways: bytes of storage, and I/O operations.  And an idle disk means
we're wasting part of its value it could be providing, and if the goal
is to decrease the overall storage TCO, wasting IOPS is not a good
thing[3].  Hence, we try to organize our data to keep all of the hard
drives busy, by peanut-buttering the hot data across all of the disks
in the cluster[4].

[3] https://research.google/pubs/pub44830/
[4] http://www.pdsw.org/pdsw-discs17/slides/PDSW-DISCS-Google-Keynote.pdf

Hence, a spun-down disk is a disk which is frittering away the CapEx
of the drive and a portion of the server cost to which the disk is
attached.  And if you can find useful work for that disk to do, it's
way more valuable to keep it spun up even taking into account to the
power and air-conditioning costs of the spinning drive.


It should also be noted that modern HDD's now also have *write*
limits[5], just like SSD's.  This is especially true for technologies
like HAMR --- where if you need to apply *heat* to write, that means
additional thermal stress on the drive head when you write to a disk,
but the write limits predate new technologies like HAMR and MAMR.

[5] https://www.theregister.com/2016/05/03/when_did_hard_drives_get_workload_rate_limits/

HDD write limits has implications for systems that are using log
structured storage, or other copy-on-write schemes, or systems that
are moving data around to balance hot and cold data as described in
the PDSW keynote.  This is probably not an issue for home systems, but
it's one of the things which keeps the storage space interesting.  :-)

					- Ted

P.S.  I have a Synology NAS box, and I *do* let the disks spin down.
Storage at the industrial scale is really different than storage at
the personal scale.  I do use RAID, but my backup strategy in extremis
is encrypted backups uploaded to cloud storage (where I can take
advantage of industrial-scale storage pricing).

  reply	other threads:[~2021-09-02 16:58 UTC|newest]

Thread overview: 21+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-08-29 22:12 Jon Steinhart
2021-08-29 23:09 ` Henry Bent
2021-08-30  3:14   ` Theodore Ts'o
2021-08-30 13:55     ` Steffen Nurpmeso
2021-08-30  9:14   ` John Dow via TUHS
2021-08-29 23:57 ` Larry McVoy
2021-08-30  1:21   ` Rob Pike
2021-08-30  3:46   ` Theodore Ts'o
2021-08-30 23:04     ` Bakul Shah
2021-09-02 15:52       ` Jon Steinhart
2021-09-02 16:57         ` Theodore Ts'o [this message]
2021-08-30  3:36 ` Bakul Shah
2021-08-30 11:56   ` Theodore Ts'o
2021-08-30 22:35     ` Bakul Shah
2021-08-30 15:05 ` Steffen Nurpmeso
2021-08-31 13:18   ` Steffen Nurpmeso
2021-08-30 21:38 ` Larry McVoy
2021-08-30 13:06 Norman Wilson
2021-08-30 14:42 ` Theodore Ts'o
2021-08-30 18:08   ` Adam Thornton
2021-08-30 16:46 ` Arthur Krewat

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=YTECjlhuKlC4aIad@mit.edu \
    --to=tytso@mit.edu \
    --cc=jon@fourwinds.com \
    --cc=tuhs@minnie.tuhs.org \
    --subject='Re: [TUHS] Is it time to resurrect the original dsw (delete with switches)?' \
    /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

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).