SandForce-based SSDs

M

Man-wai Chang

Are they really that bad?

--
@~@ You have the right to remain silence.
/ v \ Simplicity is Beauty! May the Force and farces be with you!
/( _ )\ (Fedora 15 i686) Linux 3.3.1
^ ^ 01:55:01 up 2 days 6:12 0 users load average: 0.00 0.01 0.05
ä¸å€Ÿè²¸! ä¸è©é¨™! ä¸æ´äº¤! ä¸æ‰“交! ä¸æ‰“劫! ä¸è‡ªæ®º! è«‹è€ƒæ…®ç¶œæ´ (CSSA):
http://www.swd.gov.hk/tc/index/site_pubsvc/page_socsecu/sub_addressesa
 
Y

Yousuf Khan

Are they really that bad?

Apparently, since they are the most popular brand of SSD chipset on the
planet, there is just more units to complain about. Some of their
problems get fixed with firmware upgrades. I just bought a Corsair Force
3 SSD, which has the Sandforce controller, and I did have some problems
with it.

When I started out, I was using the drives in AHCI mode, because I had
heard that these are the only drivers which have all of the support for
SSD-specific commands. Everything seemed fine, until occasionally I
would get a weird random lockup during normal operations. I would be
doing something normal in Windows and then all of a sudden a sudden and
unexpected freeze would occur in the system. I could move the mouse, but
I couldn't click on anything and every program would just sit there
unable to do anything until this freeze-up passed, which usually took
about 10 seconds only. But I'd get maybe a couple or more in a day. It
was getting annoying. Then I went on the Corsair forum, and found out
that this is one of the most common complaints about the Corsair SSD's.
I tried everything to see if I could fix it like turning off
write-caching or turning off the TRIM support. Nothing helped. Then on a
lark I tried it under IDE mode, and the freezes went away! I said, well
this isn't good, because there's no way IDE mode supports the SSD TRIM
command. Surprisingly, when I ran the command to turn on TRIM support,
it worked under the IDE driver too! So it looks like running the thing
in IDE mode is the best option: it has full support for TRIM, and it
doesn't lockup like AHCI does, and it is only slightly slower than AHCI.
My Windows Experience Index for the disk went down from 7.6 to 7.1 (and
that's out of 7.9 in both cases), which is a very minuscule difference,
not something that can be felt in human terms. I'll give up the slight
performance for greater stability.

Yousuf Khan
 
A

Arno

Apparently, since they are the most popular brand of SSD chipset on the
planet, there is just more units to complain about. Some of their
problems get fixed with firmware upgrades. I just bought a Corsair Force
3 SSD, which has the Sandforce controller, and I did have some problems
with it.
When I started out, I was using the drives in AHCI mode, because I had
heard that these are the only drivers which have all of the support for
SSD-specific commands. Everything seemed fine, until occasionally I
would get a weird random lockup during normal operations. I would be
doing something normal in Windows and then all of a sudden a sudden and
unexpected freeze would occur in the system. I could move the mouse, but
I couldn't click on anything and every program would just sit there
unable to do anything until this freeze-up passed, which usually took
about 10 seconds only. But I'd get maybe a couple or more in a day. It
was getting annoying. Then I went on the Corsair forum, and found out
that this is one of the most common complaints about the Corsair SSD's.
I tried everything to see if I could fix it like turning off
write-caching or turning off the TRIM support. Nothing helped. Then on a
lark I tried it under IDE mode, and the freezes went away! I said, well
this isn't good, because there's no way IDE mode supports the SSD TRIM
command. Surprisingly, when I ran the command to turn on TRIM support,
it worked under the IDE driver too! So it looks like running the thing
in IDE mode is the best option: it has full support for TRIM, and it
doesn't lockup like AHCI does, and it is only slightly slower than AHCI.
My Windows Experience Index for the disk went down from 7.6 to 7.1 (and
that's out of 7.9 in both cases), which is a very minuscule difference,
not something that can be felt in human terms. I'll give up the slight
performance for greater stability.
Yousuf Khan

Interesting. I have two OCZ's (128GB Vertwex 2 and 256GB Vertex 3)
and never noticed any problem. Of course these are not the
cheapest models and both were bought when they had been
on the market for a while. For new products, it is not a
surprise that problems crop up. That even happens with HDDs
occasionally and they are (or should be) well-understood
technology. My advice would be to buy main-stream (larger numbers)
SSD models that have been on the market at least half a year.
You can research that particular model before. This will
not give you useful failure probability data (vocal minority
problem) but will show you the potential failure modes
and incompatibilities. Overall, my take is that SSDs are about
as reliable as HDDs.

Arno
 
M

Man-wai Chang

technology. My advice would be to buy main-stream (larger numbers)
SSD models that have been on the market at least half a year.
You can research that particular model before. This will
not give you useful failure probability data (vocal minority
problem) but will show you the potential failure modes
and incompatibilities. Overall, my take is that SSDs are about
as reliable as HDDs.

I read that SandForce chipset would try to use compression when storing
data. Could you disable that feature?

--
@~@ You have the right to remain silence.
/ v \ Simplicity is Beauty! May the Force and farces be with you!
/( _ )\ (Fedora 15 i686) Linux 3.3.1
^ ^ 03:11:02 up 7:33 0 users load average: 0.00 0.01 0.05
ä¸å€Ÿè²¸! ä¸è©é¨™! ä¸æ´äº¤! ä¸æ‰“交! ä¸æ‰“劫! ä¸è‡ªæ®º! è«‹è€ƒæ…®ç¶œæ´ (CSSA):
http://www.swd.gov.hk/tc/index/site_pubsvc/page_socsecu/sub_addressesa
 
M

Man-wai Chang

command. Surprisingly, when I ran the command to turn on TRIM support,
it worked under the IDE driver too! So it looks like running the thing
in IDE mode is the best option: it has full support for TRIM, and it
doesn't lockup like AHCI does, and it is only slightly slower than AHCI.

But when you downgrade one port of the SATA chipset to IDE, it would
affect the speed of all other storage devices.... :)

--
@~@ You have the right to remain silence.
/ v \ Simplicity is Beauty! May the Force and farces be with you!
/( _ )\ (Fedora 15 i686) Linux 3.3.1
^ ^ 03:16:02 up 7:38 0 users load average: 0.00 0.01 0.05
ä¸å€Ÿè²¸! ä¸è©é¨™! ä¸æ´äº¤! ä¸æ‰“交! ä¸æ‰“劫! ä¸è‡ªæ®º! è«‹è€ƒæ…®ç¶œæ´ (CSSA):
http://www.swd.gov.hk/tc/index/site_pubsvc/page_socsecu/sub_addressesa
 
Y

Yousuf Khan

Interesting. I have two OCZ's (128GB Vertwex 2 and 256GB Vertex 3)
and never noticed any problem. Of course these are not the
cheapest models and both were bought when they had been
on the market for a while. For new products, it is not a
surprise that problems crop up. That even happens with HDDs
occasionally and they are (or should be) well-understood
technology. My advice would be to buy main-stream (larger numbers)
SSD models that have been on the market at least half a year.
You can research that particular model before. This will
not give you useful failure probability data (vocal minority
problem) but will show you the potential failure modes
and incompatibilities. Overall, my take is that SSDs are about
as reliable as HDDs.

The model I bought was introduced in early 2011, so it's been around 1
year already. The price cuts did make it attractive though. When doing
research, you're mainly looking at reviews, benchmarks, and optimization
advice. You rarely see problems crop up in reviews (perhaps they get the
cherry-picked units). You only notice the problems after you actually
get them yourself, and then do a search on the forums for this same
problem. The unit I have had already been flashed to the latest
firmware, and so it didn't seem like there should be any problems still,
but there was. Fortunately, I was able to discover the
solution/workaround myself.

Yousuf Khan
 
Y

Yousuf Khan

But when you downgrade one port of the SATA chipset to IDE, it would
affect the speed of all other storage devices.... :)

I've done extensive benchmarking on all of my drives, I never noticed
any difference in speed on the mechanical drives when used in AHCI or
IDE modes. And there was only a minor degradation of the speed on the
SSD. Although I had switched over the AHCI mode over a year ago, even
before I bought the SSD, it did not have any features I considered
necessary yet, but might be in the future. Such as NCQ and hot-plugging.
Neither of them came in particularly useful to me, so I don't miss them
now that I've disabled them by switching back to the IDE drivers. NCQ
doesn't work on SSD's anyways, they only matter to hard disks. TRIM
doesn't matter to hard disks, they only work on NCQ.

Yousuf Khan
 
Y

Yousuf Khan

I read that SandForce chipset would try to use compression when storing
data. Could you disable that feature?

Nope, it's built-in to the chipset and beyond the control of the OS. It
actually helps when you have compressible data, and it doesn't hurt when
you don't. This shows the relative difference in speeds between
compressible and non-compressible data:

http://www.xbitlabs.com/articles/storage/display/corsair-force-gt_3.html

You get close to a 2 fold increase with compressible data in every case.
With compressible data, you can get up close to the full SATA 3 speed
limits on these drives (around 500 MB/s), in some cases.

Yousuf Khan
 
A

Arno

The model I bought was introduced in early 2011, so it's been around 1
year already. The price cuts did make it attractive though. When doing
research, you're mainly looking at reviews, benchmarks, and optimization
advice. You rarely see problems crop up in reviews (perhaps they get the
cherry-picked units). You only notice the problems after you actually
get them yourself, and then do a search on the forums for this same
problem. The unit I have had already been flashed to the latest
firmware, and so it didn't seem like there should be any problems still,
but there was. Fortunately, I was able to discover the
solution/workaround myself.

Hmm. Bad luck indeed. But good that you could fix this problem
yourself.

Arno
 
Y

Yousuf Khan

Hmm. Bad luck indeed. But good that you could fix this problem
yourself.


I tend to get the strangest problems almost all of the time on my
systems. But I also find a lot of solutions to them, which helps out
other people later.

Yousuf Khan
 
M

Man-wai Chang

Nope, it's built-in to the chipset and beyond the control of the OS. It
actually helps when you have compressible data, and it doesn't hurt when
you don't. This shows the relative difference in speeds between
compressible and non-compressible data:

Would the CPU be involved when compressing data?
You get close to a 2 fold increase with compressible data in every case.
With compressible data, you can get up close to the full SATA 3 speed
limits on these drives (around 500 MB/s), in some cases.

Compression make it harder to recover data!!! :)

--
@~@ You have the right to remain silence.
/ v \ Simplicity is Beauty! May the Force and farces be with you!
/( _ )\ (Fedora 15 i686) Linux 3.3.1
^ ^ 21:13:01 up 1 day 1:35 0 users load average: 0.00 0.01 0.05
ä¸å€Ÿè²¸! ä¸è©é¨™! ä¸æ´äº¤! ä¸æ‰“交! ä¸æ‰“劫! ä¸è‡ªæ®º! è«‹è€ƒæ…®ç¶œæ´ (CSSA):
http://www.swd.gov.hk/tc/index/site_pubsvc/page_socsecu/sub_addressesa
 
M

Man-wai Chang

I've done extensive benchmarking on all of my drives, I never noticed
any difference in speed on the mechanical drives when used in AHCI or
IDE modes. And there was only a minor degradation of the speed on the

The speed decrease in the SATA wires are undeniable, even though an
average mortal would not sense the differences.

Maybe SSDs really should be mounted on its own SATA chipset in the
motherboard, rather than blending in with other 100% RELIABLE
traditional hard disks and DVD/BluRay drives... :)

--
@~@ You have the right to remain silence.
/ v \ Simplicity is Beauty! May the Force and farces be with you!
/( _ )\ (Fedora 15 i686) Linux 3.3.1
^ ^ 02:32:01 up 1 day 6:54 0 users load average: 0.06 0.05 0.05
ä¸å€Ÿè²¸! ä¸è©é¨™! ä¸æ´äº¤! ä¸æ‰“交! ä¸æ‰“劫! ä¸è‡ªæ®º! è«‹è€ƒæ…®ç¶œæ´ (CSSA):
http://www.swd.gov.hk/tc/index/site_pubsvc/page_socsecu/sub_addressesa
 
B

bbbl67

The speed decrease in the SATA wires are undeniable, even though an
average mortal would not sense the differences.

The differences that I notice is all that I care about. I noticed a
huge difference when going from HDD to SSD, but didn't notice a
difference going from AHCI to IDE. I'm sure it'll make a difference to
my bragging rights, but it really doesn't make a difference to overall
daily life. Booting times haven't changed, and application loading
times haven't either.
 
R

Rod Speed

Yousuf Khan wrote
Arno wrote
The model I bought was introduced in early 2011, so it's been around 1 year already. The price cuts did make it
attractive though. When doing research, you're mainly looking at reviews, benchmarks, and optimization advice. You
rarely see problems crop up in reviews (perhaps they get the cherry-picked units).

I think its more that they are careful to not put the boot in too agressively
otherwise they wont get anything more to review from that operation etc.

Even our consumer operation that deliberately buys stuff at retail in
an attempt to avoid that problem, often doesnt manage to detect
the duds that do later show up once lots are bought by consumers.
You only notice the problems after you actually get them yourself, and then do a search on the forums for this same
problem.

And presumably most of those that end up with a problem never bother
to say anything about it on any forum.
The unit I have had already been flashed to the latest firmware, and so it didn't seem like there should be any
problems still, but there was.

And that shows how unsatisfactory it is to try to work out the duds before you buy.
 
A

Arno

Would the CPU be involved when compressing data?
No.
Compression make it harder to recover data!!! :)

Indeed. But as you need to unsolder SMD chips for that,
and deal with wear-leveling and defect-management, it is
already pretty hard.

Arno
 
Y

Yousuf Khan

Would the CPU be involved when compressing data?

No, entirely done within the SSD's controller.
Compression make it harder to recover data!!! :)

Forget about all of your hard-earned hard disk knowledge, it doesn't
apply here anymore! :)

You'll never be able to recover any data that's been deleted from an
SSD. Unlike with a hard disk, once you delete a file (and subsequently
remove it from the Recycle Bin too, if that's enabled), the empty area
actually is fully erased. It's like you've done a secure shred of your
empty spaces everytime!

One thing that surprised me about SSD's when I learned about it is that
you can't simply overwrite data on the sectors like with an HDD, you
must first erase it, and then write to it. The reason is based on how
flash memory works. You can only write ones to a flash memory block, you
can't write zeros! So what they do is send an erase/reset command to the
block, which zeros all of the bits out. You then go in and write a new
set of ones to the proper bit locations, leaving other bit locations
untouched at zero. This is why they brought the TRIM command in. Because
you have to spend so much time resetting blocks, they now maintain a
pool of already erased blocks on the drive. That way when it's time to
rewrite a block, you actually end up writing to a totally different
block which replaces the old block, and the old block gets erased in the
background with TRIM. Then that old block gets put into the reserved
pool ready to be used another time.

Yousuf Khan
 
Y

Yousuf Khan

That's not entirely true - it depends on the garbage collection
algorithms, and what else is in the same erase block.

It's true that your erased data /might/ be zapped quickly - but it might
lie around for ever afterwards, even if it has been "shredded", "wiped",
"securely deleted", or otherwise "deleted".

I don't see it likely to lie around forever. If any deleted data lies
around for longer than a minute, I'd be surprised. They usually need
these blocks to go to the reserved pool as soon as possible so that they
always have a source of empty blocks to which they can write to. The
garbage collection won't kick in during moments of activity, for sure,
but I also don't see any sort of continuous fully-loaded disk activity
lasting more than a minute on an SSD. It's possible that disk activity
can last very long on an HDD, but an SSD is so fast it will not last
that long.

Hell, you can do a full secure erase of the SSD in less than a second --
the entire SSD.
It is certainly true that it is difficult to get erased data off the SSD
- you need to remove the chips and read them individually, and figure
out how all the re-location and indirection works. Even if you have
access to the re-location algorithms and data structures, it would still
be a hard task to reconstruct things.

And you'd pretty much have to pull the power plug off the SSD on a
running system in less than a minute after some fully-loaded disk
activity operation so that the garbage collection hasn't had a chance to
finish.
So if you are paranoid about hiding your old data, you can't do it with
an SSD except by a full disk secure erase. And on the other hand, if you
/want/ to recover lost data, you probably can't do it.

I don't think so, everything marked for deletion pretty much is deleted
on an SSD, since all you have to do is send the block reset command,
which just drops everything to zero on the block.

Yousuf Khan
 
Y

Yousuf Khan

Yousuf,


What's your take on these SSD's in a PC running 24/7/365, as compared
to one running std HDD's? Longevity? Reliability?

Initially, I was a little hesitant after hearing about their downsides.
Now that I know a bit more about them, all of the fears about them dying
quickly seem to be overblown, they have the art of wear-leveling down
pat. If anything, an SSD is the perfect device to be running 24/7, as
there are no moving parts, and the temperature never goes up or down
with activity.

Another thing to consider is that even after an SSD has died, it doesn't
become unusable, you can still read from it, it just becomes read-only.
So you can still get all of your left over data off of it, as long as
the controller is still intact. The controller has the same amount of
vulnerability as the controller on a hard disk.
With your large assortment of SSD's, any preferences you might
have/suggestions for newbies to SSD's?

A large assortment of one! However, after reading the forums, it looks
like there are two major controller chipsets available for SSD's: (1)
the Marvell, and (2) the Sandforce. There is also the Intel controller,
only for Intel SSD's. #2 is by far the more popular, but it uses up a
bit more memory within its reserve pool, approximately 7% usually. So
with a Marvell controller, you'll typically see sizes like
32/64/128/256GB. But with a Sandforce, you'll typically see sizes like
30/60/120/240GB. The general consensus is that the additional reserved
pool on the Sandforce is good for reliability and write performance,
since there are more reserved sectors for balancing the wear-level with,
but capacity sucks.

Yousuf Khan
 
M

Man-wai Chang

can't write zeros! So what they do is send an erase/reset command to the
block, which zeros all of the bits out. You then go in and write a new
set of ones to the proper bit locations, leaving other bit locations
untouched at zero. This is why they brought the TRIM command in.

Thanks. Didn't know about this.


--
@~@ You have the right to remain silence.
/ v \ Simplicity is Beauty! May the Force and farces be with you!
/( _ )\ (Fedora 15 i686) Linux 3.3.1
^ ^ 16:06:01 up 3 days 7:04 0 users load average: 0.01 0.05 0.05
ä¸å€Ÿè²¸! ä¸è©é¨™! ä¸æ´äº¤! ä¸æ‰“交! ä¸æ‰“劫! ä¸è‡ªæ®º! è«‹è€ƒæ…®ç¶œæ´ (CSSA):
http://www.swd.gov.hk/tc/index/site_pubsvc/page_socsecu/sub_addressesa
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Top