Installing OpenBSD on SSD drives

[Jeff_1981]

 Hello,
Is there any particular problem with installing OpenBSD on a SSD HD ?  I
once could on one machine but on my actual machine it simply does'nt work.
After a while, the SSD disk becomes like overloaded and unavailable to
continue the installing process of 4.6.
Regards

[STeve Andre']

On Wednesday 04 November 2009 16:10:06 Jean-Frangois SIMON wrote:
>  Hello,
> Is there any particular problem with installing OpenBSD on a SSD HD ?  I
> once could on one machine but on my actual machine it simply does'nt work.
> After a while, the SSD disk becomes like overloaded and unavailable to
> continue the installing process of 4.6.
> Regards

I played with one, briefly, and it seemed to work.  A litte weird, not
hearing anything from it...

But I'm not at all eager to actually use them just yet.  Look for the
goofs Intel has had with them.  How long will they last, and what is
the failure mode like?  More often than not a spinning disk will give
notice of impending death with a few bad spots before The End.  But
what of an SSD?  By its very nature I could see an address line going,
leaving a very weird pattern of unaffected data.

SSDs are the future, I'm fairly sure but I think they need to mature
as well as get bigger.

Lastly, saying where the install hangs would really help.  And of
course how big is it and who made it?

--STeve Andre'

[Roger Schreiter]

Jean-Frangois SIMON schrieb:
> ...
> Is there any particular problem with installing OpenBSD on a SSD HD ?  I

Hello,

it is like for any OS on SSD HD. Make sure, you are using
no swap partition!

And if you are using an application, which is writing
a lot of things into files, put the respective dirs into
ramdisks!

We are running some embedded PCs with OpenBSD, which have the
SSD HD completely write protected. All partitions are
mounted read only, and /tmp, /dev and /var is put into
ramdisks. Works fine.

Regards,
Roger.

[Ted Unangst-2]

2009/11/4 Roger Schreiter <roger@...>:
> it is like for any OS on SSD HD. Make sure, you are using
> no swap partition!

This is ridiculous advice.

> And if you are using an application, which is writing
> a lot of things into files, put the respective dirs into
> ramdisks!

Combined with this is even dumber.

If you can't swap, you're already in trouble if you run into memory
pressure.  So then you go and put the filesystem in RAM to make sure
there's lots of extra memory pressure?

[Aaron Mason]

On Thu, Nov 5, 2009 at 9:12 AM, Ted Unangst <ted.unangst@...> wrote:

> 2009/11/4 Roger Schreiter <roger@...>:
>> it is like for any OS on SSD HD. Make sure, you are using
>> no swap partition!
>
> This is ridiculous advice.
>
>> And if you are using an application, which is writing
>> a lot of things into files, put the respective dirs into
>> ramdisks!
>
> Combined with this is even dumber.
>
> If you can't swap, you're already in trouble if you run into memory
> pressure.  So then you go and put the filesystem in RAM to make sure
> there's lots of extra memory pressure?
>
>

I'm with Ted on this one.  At the very least, stick a USB drive in and
use that for swap.  If things are going to write to SSDs a lot, get
two (if budget allows) and stripe/RAID-5 them - this actually does
wonders for increasing the lifespan of SSDs.

--
Aaron Mason - Programmer, open source addict
- Oh, why does everything I whip leave me?

[philippe aubry]

Hello,
I'm using a 32 GB SSD drive from approximatly one year with openBSD 4.4 into
a SOEKRIS and no troubles with that, the great think is NO NOISE, NO HEAT.
I used the soekris as firewall and the uptime is approximatly 178 days.

Regards

2009/11/4 Jean-Frangois SIMON <jfsimon1981@...>

>  Hello,
> Is there any particular problem with installing OpenBSD on a SSD HD ?  I
> once could on one machine but on my actual machine it simply does'nt work.
> After a while, the SSD disk becomes like overloaded and unavailable to
> continue the installing process of 4.6.
> Regards

[Aaron Mason]

2009/11/5 Jean-Frangois SIMON <jfsimon1981@...>:
>  Hello,
> Is there any particular problem with installing OpenBSD on a SSD HD ?  I
> once could on one machine but on my actual machine it simply does'nt work.
> After a while, the SSD disk becomes like overloaded and unavailable to
> continue the installing process of 4.6.
> Regards
>
>

Hi Jean-Francios,

Is this a used SSD?  That happens with used ones because they end up
doing twice the work - once to erase the used block and again to
actually write the block (and several blocks around them, AAMOF).

If you have a "secure erase" option available, use it.  This will
restore the data blocks to an unused state, and restore full speed
again.

HTH

2009/11/5 STeve Andre' <andres@...>
>But I'm not at all eager to actually use them just yet.  Look for the
>goofs Intel has had with them.  How long will they last, and what is
>the failure mode like?  More often than not a spinning disk will give
>notice of impending death with a few bad spots before The End.  But
>what of an SSD?  By its very nature I could see an address line going,
>leaving a very weird pattern of unaffected data.

I'd say SMART would answer the call by sending DANGER WILL ROBINSON
messages to the OS - it would be up to the OS to intercept these
messages and inform the sysadmin, however.

My $0.02.

--
Aaron Mason - Programmer, open source addict
- Oh, why does everything I whip leave me?

[K Kadow]

2009/11/4 Jean-Frangois SIMON <jfsimon1981@...>:
>  Hello,
> Is there any particular problem with installing OpenBSD on a SSD HD ?  I
> once could on one machine but on my actual machine it simply does'nt work.
> After a while, the SSD disk becomes like overloaded and unavailable to
> continue the installing process of 4.6.
> Regards

Sounds like an issue with your SSD?
Can you supply a dmesg, and details on the SSD, make/model/supplier,
as well as the motherboard and how the drive appears to the BIOS?


On Wed, Nov 4, 2009 at 4:12 PM, Ted Unangst <ted.unangst@...> wrote:
> 2009/11/4 Roger Schreiter <roger@...>:
>> it is like for any OS on SSD HD. Make sure, you are using
>> no swap partition!
>
> This is ridiculous advice.

This *was* reasonable advice for the older generations of
CompactFlash, but may no longer be a consideration with newer
flash/SSD drives.

I have run many embedded servers (mostly OpenBSD on Soekris) without
swap, never had any problems traceable to the lack of swap space.


>> And if you are using an application, which is writing
>> a lot of things into files, put the respective dirs into
>> ramdisks!
>
> Combined with this is even dumber.
>
> If you can't swap, you're already in trouble if you run into memory
> pressure.  So then you go and put the filesystem in RAM to make sure
> there's lots of extra memory pressure?

Actually, the above is standard advice for running any Unix on flash,
as people have been doing with Soekris and CF since at least 2001.

The idea isn't to put "the filesystem" into RAM, but rather to reduce
the write operations by mounting filesystems used for frequently
written smal files (e.g. /var/tmp) as ramdisks.

Kevin

[Robert-101]

On Wed, 04 Nov 2009 23:00:39 +0100
Roger Schreiter <roger@...> wrote:

> Jean-Frangois SIMON schrieb:
> > ...
> > Is there any particular problem with installing OpenBSD on a SSD
> > HD ?  I
>
> Hello,
>
> it is like for any OS on SSD HD. Make sure, you are using
> no swap partition!
>
> And if you are using an application, which is writing
> a lot of things into files, put the respective dirs into
> ramdisks!
>
> We are running some embedded PCs with OpenBSD, which have the
> SSD HD completely write protected. All partitions are
> mounted read only, and /tmp, /dev and /var is put into
> ramdisks. Works fine.
>
> Regards,
> Roger.

That advice might have had some merit with 1GB Compact Flash drives ...

On eg. a 80GB SSD partition 60 and leave the rest empty. With that you
have _a lot_ of sectors to remap in case some fail. That will increase
the lifetime of the drive.
Usually flash fales gracefully, can't write but still read, so one would
be able to recover the data.
Flash is no mirical cure, having backups is still mandatory.
I don't expect my 2,5" drive in my laptop to last longer than the
stated 5 years the avarage MLC SSD gets quoted. All that banging around,
even turned off, in the laptop bag takes it's toll.
Harddrives that store critical data are swapped in the 2 to 3 year time
frame at latest, if they didn't fail on their own before and are
repurposed in less crucial systems like desktops. (...less potential
downtime, less power consumption, more peace of mind)

On the gp's topic, there is nothing special about SSD's that should
keep them from working like any other (guessing) SATA device.
("It doesn't work!" Isn't anywhere near a cry for help that warrants an
answer...)

- Robert

[Roger Schreiter]

Ted Unangst schrieb:
> ...
>> no swap partition!
>
> This is ridiculous advice.
> ...
>> a lot of things into files, put the respective dirs into
>> ramdisks!
>
> Combined with this is even dumber.


Hi,

anyway, intensive swapping onto SDD HD will destroy your SDD HD.

If RAM is the limiting resource in your system, you are right,
my advice is ridiculous.

In any else case, my advice is important, and for many, many
applications it is possible to equip a system with enough RAM,
making swapping uneccessary.

Ramdisks and complete write protections of the HD is of course
just an option to think about, and depends on the application,
if appropriate or not.



Regards,
Roger.

[Ted Unangst-2]

On Wed, Nov 4, 2009 at 5:44 PM, K K <kkadow@...> wrote:
> This *was* reasonable advice for the older generations of
> CompactFlash, but may no longer be a consideration with newer
> flash/SSD drives.
>
> I have run many embedded servers (mostly OpenBSD on Soekris) without
> swap, never had any problems traceable to the lack of swap space.

Why do we keep repeating 10 year old advice that may have applied to
crappy 1GB flash and think it matters for 100GB drives using rather
different technology?  When you run newfs, do you make sure to line
the cylinder groups up just right?  Because that was standard advice
too.

More relevantly, I bet you never tried starting firefox on your
soekris.  Why does everyone assume that the only possible use for an
SSD hard drive is in some crippled embedded box?  Heaven forbid
somebody put a fast drive in a computer that they'll actually use.

I've got 4GB of SSD swap in my laptop.  Yes, 4GB of swap on SSD!
OMG!!!  It'll wear out in a month!

Why do you assume that merely creating a swap partition somehow forces
the kernel to use it?  If your system is running without a swap
partition, it can run without writing to swap too.

[Otto Moerbeek]

On Wed, Nov 04, 2009 at 11:52:53PM +0100, Roger Schreiter wrote:

> Ted Unangst schrieb:
> > ...
> >> no swap partition!
> >
> > This is ridiculous advice.
> > ...
> >> a lot of things into files, put the respective dirs into
> >> ramdisks!
> >
> > Combined with this is even dumber.
>
>
> Hi,
>
> anyway, intensive swapping onto SDD HD will destroy your SDD HD.

Nonsense. modern ssd have firmware to avoid the wear out problem.
Yes, sometimes the firmware is buggy. That is a bigger problem than
the wearout.

>
> If RAM is the limiting resource in your system, you are right,
> my advice is ridiculous.
>
> In any else case, my advice is important, and for many, many
> applications it is possible to equip a system with enough RAM,
> making swapping uneccessary.
>
> Ramdisks and complete write protections of the HD is of course
> just an option to think about, and depends on the application,
> if appropriate or not.
>
>
>
> Regards,
> Roger.

[Ted Unangst-2]

On Wed, Nov 4, 2009 at 5:52 PM, Roger Schreiter <roger@...> wrote:
> anyway, intensive swapping onto SDD HD will destroy your SDD HD.

I just reread this and noticed the use of the word intensive.  As if
there's no middle ground between "no swap" and "intensive swapping".

Again, the mere presence of a swap partition will not cause the kernel
to just randomly splat gobs of data onto disk.  There is no harm in
having a swap partition that you don't need, and it's a lot better
than needing swap and not having it.

[Joachim Schipper-2]

On Thu, Nov 05, 2009 at 09:39:08AM +1100, Aaron Mason wrote:

> 2009/11/5 STeve Andre' <andres@...>
> >But I'm not at all eager to actually use [SSDs] just yet.  Look for the
> >goofs Intel has had with them.  How long will they last, and what is
> >the failure mode like?  More often than not a spinning disk will give
> >notice of impending death with a few bad spots before The End.  But
> >what of an SSD?  By its very nature I could see an address line going,
> >leaving a very weird pattern of unaffected data.
>
> I'd say SMART would answer the call by sending DANGER WILL ROBINSON
> messages to the OS - it would be up to the OS to intercept these
> messages and inform the sysadmin, however.

Put something like the following in root's crontab:

12 * * * * /sbin/atactl wd0 smartstatus >/dev/null

That said, I'm not sure that SSDs actually send SMART messages. (OTOH,
one can't be sure with old-fashioned hard disks, either...)

                Joachim

[Joachim Schipper-2]

On Wed, Nov 04, 2009 at 11:45:15PM +0100, Robert wrote:

> On Wed, 04 Nov 2009 23:00:39 +0100
> Roger Schreiter <roger@...> wrote:
> > [Installing OpenBSD on an SSD] is like for any OS on SSD HD. Make
> > sure, you are using no swap partition!
> >
> > And if you are using an application, which is writing
> > a lot of things into files, put the respective dirs into
> > ramdisks!
> >
> > We are running some embedded PCs with OpenBSD, which have the
> > SSD HD completely write protected. All partitions are
> > mounted read only, and /tmp, /dev and /var is put into
> > ramdisks. Works fine.
> >
> > Regards,
> > Roger.
>
> That advice might have had some merit with 1GB Compact Flash drives ...
>
> On eg. a 80GB SSD partition 60 and leave the rest empty. With that you
> have _a lot_ of sectors to remap in case some fail. That will increase
> the lifetime of the drive.

This seems predicated on the firmware being smart enough to swap out bad
sectors for good setors that are addressable but not used in practice.
Is the firmware that smart? (I know about wear-levelling and swapping
in "reserve" sectors, but that's different - those *cannot* be
addressed.)

                Joachim

[Henning Brauer]

* Jean-Frangois SIMON <jfsimon1981@...> [2009-11-04 22:16]:
> Is there any particular problem with installing OpenBSD on a SSD HD ?  I
> once could on one machine but on my actual machine it simply does'nt work.
> After a while, the SSD disk becomes like overloaded and unavailable to
> continue the installing process of 4.6.

no problem in general...

sd0 at scsibus0 targ 0 lun 0: <ATA, SAMSUNG MMCRE64G, VBM1> SCSI3 0/direct fixed
sd0: 61057MB, 512 bytes/sec, 125045424 sec total


--
Henning Brauer, hb@..., henning@...
BS Web Services, http://bsws.de
Full-Service ISP - Secure Hosting, Mail and DNS Services
Dedicated Servers, Rootservers, Application Hosting

[Jan Stary]

> * Jean-Frangois SIMON <jfsimon1981@...> [2009-11-04 22:16]:
> > Is there any particular problem with installing OpenBSD on a SSD HD ?

No

[Cesare Gargano-4]

On Thu, Nov 05, 2009 at 04:32:20PM +0100, Henning Brauer wrote:

> * Jean-Frangois SIMON <jfsimon1981@...> [2009-11-04 22:16]:
> > Is there any particular problem with installing OpenBSD on a SSD HD ?  I
> > once could on one machine but on my actual machine it simply does'nt work.
> > After a while, the SSD disk becomes like overloaded and unavailable to
> > continue the installing process of 4.6.
>
> no problem in general...
>
> sd0 at scsibus0 targ 0 lun 0: <ATA, SAMSUNG MMCRE64G, VBM1> SCSI3 0/direct fixed
> sd0: 61057MB, 512 bytes/sec, 125045424 sec total
>
>
> --
> Henning Brauer, hb@..., henning@...
> BS Web Services, http://bsws.de
> Full-Service ISP - Secure Hosting, Mail and DNS Services
> Dedicated Servers, Rootservers, Application Hosting

same for me

wd0 at pciide0 channel 1 drive 0: <CORSAIR CMFSSD-128GBG1D>
...

[Ted Unangst-2]

On Thu, Nov 5, 2009 at 10:28 AM, Joachim Schipper
<joachim@...> wrote:
> This seems predicated on the firmware being smart enough to swap out bad
> sectors for good setors that are addressable but not used in practice.
> Is the firmware that smart? (I know about wear-levelling and swapping
> in "reserve" sectors, but that's different - those *cannot* be
> addressed.)

There are no reserve sectors, there's just sectors.  Some of them are
reserved, but they're no different from the normal sectors.

Think of it like a 6GB machine running PAE (and only one process).
You can only address 4GB at maximum, but if something goes bad,
there's other memory your virtual addresses can get mapped to.  If you
are only writing to 1GB of space though, it's easily spread out over
all 6GB.  The high 2GB is not special or different.

[Otto Moerbeek]

On Thu, Nov 05, 2009 at 12:14:15PM -0500, Ted Unangst wrote:

> On Thu, Nov 5, 2009 at 10:28 AM, Joachim Schipper
> <joachim@...> wrote:
> > This seems predicated on the firmware being smart enough to swap out bad
> > sectors for good setors that are addressable but not used in practice.
> > Is the firmware that smart? (I know about wear-levelling and swapping
> > in "reserve" sectors, but that's different - those *cannot* be
> > addressed.)
>
> There are no reserve sectors, there's just sectors.  Some of them are
> reserved, but they're no different from the normal sectors.
>
> Think of it like a 6GB machine running PAE (and only one process).
> You can only address 4GB at maximum, but if something goes bad,
> there's other memory your virtual addresses can get mapped to.  If you
> are only writing to 1GB of space though, it's easily spread out over
> all 6GB.  The high 2GB is not special or different.

This article gives some background info:
http://arstechnica.com/hardware/news/2009/11/biography-solid-state-disk.ars

        -Otto