Discussion:
TRIM on SSD
Add Reply
Rupert Gallagher
2017-12-05 07:26:43 UTC
Reply
Permalink
Raw Message
A production obsd is serving 50GB worth of NFS shares and hourly backups on two ssds since August, and is still going strong at 550MBps over measured 550--950Mbps LAN links. The same boots and runs the OS from a pSLC SD with Phison controller. The ssds have a 5year warrantee, and we are doing our best to stay within specs. The only concern is the lack of TRIM on the SD and the NFS ssd. With TRIM, the os keeps writing on free space instead of deleting and overwriting, for faster writing and uniform wearing of disk.

Can we safely enable TRIM on 6.2 now?

Sent from ProtonMail Mo
Juan Francisco Cantero Hurtado
2017-12-05 15:29:09 UTC
Reply
Permalink
Raw Message
Post by Rupert Gallagher
A production obsd is serving 50GB worth of NFS shares and hourly backups on two ssds since August, and is still going strong at 550MBps over measured 550--950Mbps LAN links. The same boots and runs the OS from a pSLC SD with Phison controller. The ssds have a 5year warrantee, and we are doing our best to stay within specs. The only concern is the lack of TRIM on the SD and the NFS ssd. With TRIM, the os keeps writing on free space instead of deleting and overwriting, for faster writing and uniform wearing of disk.
Can we safely enable TRIM on 6.2 now?
OpenBSD doesn't support TRIM.
--
Juan Francisco Cantero Hurtado http://juanfra.info
Rupert Gallagher
2017-12-05 22:26:27 UTC
Reply
Permalink
Raw Message
Why?

Sent from ProtonMail Mobile
On Tue, Dec 05, 2017 at 02:26:43AM -0500, Rupert Gallagher wrote: > A production obsd is serving 50GB worth of NFS shares and hourly backups on two ssds since August, and is still going strong at 550MBps over measured 550--950Mbps LAN links. The same boots and runs the OS from a pSLC SD with Phison controller. The ssds have a 5year warrantee, and we are doing our best to stay within specs. The only concern is the lack of TRIM on the SD and the NFS ssd. With TRIM, the os keeps writing on free space instead of deleting and overwriting, for faster writing and uniform wearing of disk. > > Can we safely enable TRIM on 6.2 now? OpenBSD doesn't support TRIM. -- Juan Francisco Cantero
Mike Burns
2017-12-05 23:25:13 UTC
Reply
Permalink
Raw Message
Post by Rupert Gallagher
Sent from ProtonMail Mobile
When you implement the patch that adds TRIM you might want to build off
the work already done and lessons learned. I only spent a few seconds
searching so you might find more:

https://www.tedunangst.com/flak/post/lessons-learned-about-TRIM
https://marc.info/?l=openbsd-misc&m=144738028409142&w=2
https://marc.info/?l=openbsd-misc&m=143453951901714&w=2
https://marc.info/?l=openbsd-misc&m=148908000202884&w=2
Rupert Gallagher
2017-12-06 08:15:57 UTC
Reply
Permalink
Raw Message
I know well that article, because it is several years old with no updates.

Those working on ffs should do what they are supposed to do. Lack of money? Setup a stickers sale or a kickstarter, get the money and just fucking do it.

Sent from ProtonMail Mobile
On 2017-12-05 17.26.27 -0500, Rupert Gallagher wrote: > Sent from ProtonMail Mobile When you implement the patch that adds TRIM you might want to build off the work already done and lessons learned. I only spent a few seconds searching so you might find more: https://www.tedunangst.com/flak/post/lessons-learned-about-TRIM https://marc.info/?l=openbsd-misc&m=144738028409142&w=2 https://marc.info/?l=openbsd-misc&m=143453951901714&w=2 https://marc.info/?l=openbsd-misc
Paul de Weerd
2017-12-06 08:38:02 UTC
Reply
Permalink
Raw Message
On Wed, Dec 06, 2017 at 03:15:57AM -0500, Rupert Gallagher wrote:
| I know well that article, because it is several years old with no updates.
|
| Those working on ffs should do what they are supposed to do. Lack of
| money? Setup a stickers sale or a kickstarter, get the money and
| just fucking do it.

You must be new to Open Source.

Would be funny if it wasn't so insulting.

Why don't you "just fucking do it" yourself? Apparently you'll "get
the money". The benefit for us would be that all the time you spend
working on it, you wouldn't be sending e-mails to this list.

Paul 'WEiRD' de Weerd
--
++++++++[<++++++++++>-]<+++++++.>+++[<------>-]<.>+++[<+
+++++++++++>-]<.>++[<------------>-]<+.--------------.[-]
http://www.weirdnet.nl/
Steve Litt
2017-12-07 06:21:37 UTC
Reply
Permalink
Raw Message
On Wed, 6 Dec 2017 09:38:02 +0100
Post by Paul de Weerd
| I know well that article, because it is several years old with no
updates. |
| Those working on ffs should do what they are supposed to do. Lack of
| money? Setup a stickers sale or a kickstarter, get the money and
| just fucking do it.
You must be new to Open Source.
Would be funny if it wasn't so insulting.
I just piped him to /dev/null. Besides his repeated profanity, his
telling a Free Software project what to do and how to do it, his
"who needs more than one line" formatting, his using his hatred of
archives and his possesion of an iPhone as an excuse for his hostile
email formatting, and his top-posting with vague words like "that
article" (when four were mentioned), he's just plain unhappiness
attracting.

If everybody piped him to /dev/null, nobody would be confronted with
his 1 line, 1000 word verbiage in quoted text, his useless profanity,
or his disrespect of a great Free Software project.

SteveT

Steve Litt
December 2017 featured book: Thriving in Tough Times
http://www.troubleshooters.com/thrive
Florian Obser
2017-12-06 12:19:04 UTC
Reply
Permalink
Raw Message
Post by Rupert Gallagher
I know well that article, because it is several years old with no updates.
Those working on ffs should do what they are supposed to do. Lack of money? Setup a stickers sale or a kickstarter, get the money and just fucking do it.
No.
--
I'm not entirely sure you are real.
Paulm
2017-12-06 11:56:01 UTC
Reply
Permalink
Raw Message
Post by Rupert Gallagher
I know well that article, because it is several years old with no updates.
Those working on ffs should do what they are supposed to do. Lack of
money? Setup a stickers sale or a kickstarter, get the money and just
fucking do it.
Why do you think anyone on this list should take you seriously?

You blatantly disregard the standards of this list.

Your technical observations are of highly dubious merit.

You like to make judgements/pronouncements about technical matters as
if you're an expert and can speak with authority; yet whenever you've
neeeded help for a genuine technical problem, you demonstrate that you
possess only rudimentary knowledge, at best equivalent to a
slow-learning, junior sysadmin.

And in spite of all that, you think you know best what other people
should do.

It seems you would like to think that it's your choice of email client
that engenders the aggression towards you. You're being clueless.
t***@openmailbox.org
2017-12-06 04:10:03 UTC
Reply
Permalink
Raw Message
Post by Mike Burns
Post by Rupert Gallagher
Sent from ProtonMail Mobile
When you implement the patch that adds TRIM you might want to build off
the work already done and lessons learned. I only spent a few seconds
https://www.tedunangst.com/flak/post/lessons-learned-about-TRIM
https://marc.info/?l=openbsd-misc&m=144738028409142&w=2
https://marc.info/?l=openbsd-misc&m=143453951901714&w=2
https://marc.info/?l=openbsd-misc&m=148908000202884&w=2
Oh, that's an interesting article by Unangst.

If TRIM would be implemented someday, one thing that would be neat would be that crypto and other softraid would propagate the TRIM. That would be a nice combination between wear level resiliency and disk data safety.

AFAIK some other disk crypto solutions do that. The feature comes under n
Christian Weisgerber
2017-12-06 21:31:40 UTC
Reply
Permalink
Raw Message
Post by t***@openmailbox.org
If TRIM would be implemented someday, one thing that would be
neat would be that crypto and other softraid would propagate the
TRIM. That would be a nice combination between wear level resiliency
and disk data safety.
That runs counter to popular thinking that an encrypted filesystem
should not reveal how much space has been allocated (and where).
For instance, the FAQ
https://www.openbsd.org/faq/faq14.html#softraid
explicitly recommends to overwrite the whole drive with random
data first to prevent this sort of analysis.
--
Christian "naddy" Weisgerber ***@mips.inka.de
Marc Espie
2017-12-06 22:22:53 UTC
Reply
Permalink
Raw Message
Post by Christian Weisgerber
Post by t***@openmailbox.org
If TRIM would be implemented someday, one thing that would be
neat would be that crypto and other softraid would propagate the
TRIM. That would be a nice combination between wear level resiliency
and disk data safety.
That runs counter to popular thinking that an encrypted filesystem
should not reveal how much space has been allocated (and where).
For instance, the FAQ
https://www.openbsd.org/faq/faq14.html#softraid
explicitly recommends to overwrite the whole drive with random
data first to prevent this sort of analysis.
There's an important assumption, that people actually give a fuck how much
data has been allocated.

Personally, I wouldn't.
Rupert Gallagher
2017-12-06 23:26:17 UTC
Reply
Permalink
Raw Message
Support of TRIM by ...

NetBSD, since 2012
Dragonfly BSD, since 2011
FreeBSD, since 2010
Linux, since 2008

OpenBSD, no.

Why?

Because no.

But why?

Because no. Just no. We like to say no. No, no, no. Ha ha, no!

Rise funds and do it!

No.

But why?

Because no.

Sent from ProtonMail Mobile
Post by Rupert Gallagher
A production obsd is serving 50GB worth of NFS shares and hourly backups on two ssds since August, and is still going strong at 550MBps over measured 550--950Mbps LAN links. The same boots and runs the OS from a pSLC SD with Phison controller. The ssds have a 5year warrantee, and we are doing our best to stay within specs. The only concern is the lack of TRIM on the SD and the NFS ssd. With TRIM, the os keeps writing on free space instead of deleting and overwriting, for faster writing and uniform wearing of disk.
Can we safely enable TRIM on 6.2 now?
Sen
Theo de Raadt
2017-12-07 00:22:06 UTC
Reply
Permalink
Raw Message
Post by Christian Weisgerber
Post by t***@openmailbox.org
If TRIM would be implemented someday, one thing that would be
neat would be that crypto and other softraid would propagate the
TRIM. That would be a nice combination between wear level resiliency
and disk data safety.
That runs counter to popular thinking that an encrypted filesystem
should not reveal how much space has been allocated (and where).
For instance, the FAQ
https://www.openbsd.org/faq/faq14.html#softraid
explicitly recommends to overwrite the whole drive with random
data first to prevent this sort of analysis.
I'm somewhat confused.

Naddy, did his diff get dropped when you reply to the mail?

I don't see a diff.

Maybe it was all talk, wishes for fishes, etc? If so, how
boring.
Rupert Gallagher
2017-12-08 08:07:14 UTC
Reply
Permalink
Raw Message
- UFS2 on FreeBSD supports TRIM.
- OpenBSD supports UFS2.

Is anybody using UFS2 with TRIM on OpenBSD?

Sent from ProtonMail Mobile
A production obsd is serving 50GB worth of NFS shares and hourly backups on two ssds since August, and is still going strong at 550MBps over measured 550--950Mbps LAN links. The same boots and runs the OS from a pSLC SD with Phison controller. The ssds have a 5year warrantee, and we are doing our best to stay within specs. The only concern is the lack of TRIM on the SD and the NFS ssd. With TRIM, the os keeps writing on free space instead of deleting and overwriting, for faster writing and uniform wearing of disk. Can we safely enable TRIM on 6.2
Kevin Chadwick
2017-12-08 11:03:46 UTC
Reply
Permalink
Raw Message
On Fri, 08 Dec 2017 03:07:14 -0500
Post by Rupert Gallagher
- UFS2 on FreeBSD supports TRIM.
- OpenBSD supports UFS2.
Is anybody using UFS2 with TRIM on OpenBSD?
Have you considered using a high speed HDD or RAID. From the little
information given, your performance requirements don't seem to be that
high?
Rupert Gallagher
2017-12-08 12:26:09 UTC
Reply
Permalink
Raw Message
I think you mean those round things with moving heads in a chassis with a breathing hole. No, they are not resilient to our environment.

Sent from ProtonMail Mobile
On Fri, 08 Dec 2017 03:07:14 -0500 > - UFS2 on FreeBSD supports TRIM. > - OpenBSD supports UFS2. > > Is anybody using UFS2 with TRIM on OpenBSD? Have you considered using a high speed HDD or RAID. From the little information given, your performance re
Kevin Chadwick
2017-12-08 13:09:00 UTC
Reply
Permalink
Raw Message
On Fri, 08 Dec 2017 07:26:09 -0500
Post by Rupert Gallagher
I think you mean those round things with moving heads in a chassis
with a breathing hole. No, they are not resilient to our environment.
I doubt that, we used to put them in police cars and they were fine. We
did get special ones at three times the price and mount them specially
though.

Also if they do fail you are almost guaranteed to be *able* to get the
data back which is less true of SSD.
Rupert Gallagher
2017-12-08 17:42:27 UTC
Reply
Permalink
Raw Message
They still need air, and you give it to them. We sub the server on liquid...

Sent from ProtonMail Mobile
On Fri, 08 Dec 2017 07:26:09 -0500 > I think you mean those round things with moving heads in a chassis > with a breathing hole. No, they are not resilient to our environment. I doubt that, we used to put them in police cars and they were fine. We did get special ones at three times the price and mount them specially though. Also if they do fail you are almost guaranteed to be *able* to get the dat
Rupert Gallagher
2017-12-08 17:57:35 UTC
Reply
Permalink
Raw Message
https://store.steoil.com/mineral-oil-pc-kit/

Sent from ProtonMail Mobile
They still need air, and you give it to them. We sub the server on liquid... Sent from ProtonMail Mobile On Fri, Dec 8, 2017 at 14:09, Kevin Chadwick wrote: > On Fri, 08 Dec 2017 07:26:09 -0500 > I think you mean those round things with moving heads in a chassis > with a breathing hole. No, they are not resilient to our environment. I doubt that, we used to put them in police cars and they were fine. We did get special ones at three times the price and mount them specially though. Also if they do fail you are almost guaranteed to be *able* to get the data back
Loading...