Aces High Bulletin Board

General Forums => Hardware and Software => Topic started by: MrRiplEy[H] on November 27, 2013, 09:51:00 AM

Title: SSD endurance test
Post by: MrRiplEy[H] on November 27, 2013, 09:51:00 AM
Many naysayers here are spreading FUD about SSD:s, especially the cheap TLC models.

In a heated argument about the claimed short liveness of the Samsung EVO series I dug up an endurance experiment that was an ongoing project as you can only write and erase so much data in so much time. Sadly it lacks the exact model but it does have a TLC based 840 on it.

So how has the experiment gone? As expected the TLC based Samsung died a horrible dead long time ago while... no wait! What's going on? It's still working and still as fast! The Intel 330 which was rated only for a 22Tb write life has survived for 300 Tb and has virtually no dead cells? Even the cheapest TLC is still running great after 300Tb.

http://techreport.com/review/25681/the-ssd-endurance-experiment-testing-data-retention-at-300tb

Quote
If we take my personal usage patterns as an example and use 10X write amplification as a worst-case scenario, it would take nearly 35 years to write 300TB to the flash.
Title: Re: SSD endurance test
Post by: Debrody on November 27, 2013, 10:11:00 AM
Idk... considering, the SSD uses the same technology as the RAM... RAM isnt famous for dieing after 100 rewrites.
Title: Re: SSD endurance test
Post by: MrRiplEy[H] on November 27, 2013, 11:33:42 AM
Idk... considering, the SSD uses the same technology as the RAM... RAM isnt famous for dieing after 100 rewrites.

Ram and flash are totally different tech. Ram is volatile and flash is non volatile.
Title: Re: SSD endurance test
Post by: Debrody on November 27, 2013, 01:41:30 PM
My bad, thanks.
Title: Re: SSD endurance test
Post by: guncrasher on November 27, 2013, 02:07:01 PM
cool that the ssd is performing above specs.  but will it work with every new mobo?  I keep seeing reviews about conflict with some and I can recall you mentioning something about controllers.  I have also seeing some about not working perfectly with amd in some cases.

what is it that we should look out for as compatibility goes?


semp
Title: Re: SSD endurance test
Post by: SirNuke on November 27, 2013, 02:09:17 PM
I partly read this (I'll keep for work) but I didn't see mention of how much time it took to reach 300TB?
Title: Re: SSD endurance test
Post by: gyrene81 on November 27, 2013, 02:27:18 PM
I partly read this (I'll keep for work) but I didn't see mention of how much time it took to reach 300TB?
doesn't look like the typical home desktop user or power user would be hitting that mark very quickly...assuming they aren't doing a lot with torrents and large video editing.

Quote
To be fair, our endurance experiment has lower write amplification than typical client workloads. Anvil's test is comprised almost entirely of sequential writes, while real-world desktop activity involves a lot of random I/O. There isn't a whole lot of data on the typical write amplification for client workloads, but everything I've seen and heard from SSD makers suggests a multiplication factor below 10X. If we take my personal usage patterns as an example and use 10X write amplification as a worst-case scenario, it would take nearly 35 years to write 300TB to the flash.

considering the testing methods used, most computer users would be happy with a tlc based ssd for a long time. tlc nand may not be ready for heavy enterprise level usage yet but, if manufacturers continue to over overprovision spare space to the same degree on terabyte drives as they are for those 240gb drives, ya just never know.
Title: Re: SSD endurance test
Post by: Pudgie on November 27, 2013, 02:54:10 PM
So far the 2 SSD's that I have running in my box have run flawlessly for over a year (installed them in my box on 10-'12).

They are OCZ Vertex4 256Gb SATA III MLC SSD's

They are set up w/ the 1st SSD having the OS, apps, drivers loaded on it & the 2nd SSD is being used for data storage & has the page file set up on it.
I don't do any video editing or movie stuff or CAD or tasks of the nature that typically calls for a ton of writes.

I mostly fly AHII, surf & listen to Pandora on this box w/ AHII doing most/all of the writes to the SSD's.

This is my way of testing these out.................so far they're cherry. Was looking to get a small SSD to use for page file only to really test them........may still do it.

My Asus ROG Series Rampage IV Gene mobo had no issues recognizing them, even after 2 BIOS upgrades.

My experience thus far....................

 :salute
Title: Re: SSD endurance test
Post by: Gman on November 27, 2013, 06:55:08 PM
I've read here someplace that Skuzzy/HTC didn't recommend running AH on an SSD due to all the small writes being bad for the drive.  I had already been running AH on a Samsung 830 128 gb drive for about 6 months, and it's been about 6 months since then, and it's still alive and kicking.  The drive is cheap enough that if it caks I'll just replace it with a bigger/faster/newer one.  I should get around to moving AH to my other regular spinner Caviar/Black drive I suppose, just in case, as my OS is on the SSD as well, and it would mean a system wide reformat most likely if it died.

I've been really lucky, I've never had a component go "bad" on me before I've gotten rid of it, not since the days of the Commodore 64, and that's a LOT of computers.  I blew out a joystick port once by wiping the screen of my old school monitor while holding the stick, and the electric static shock blew the port, but that was my stupidity, and not the fault of the product.

Are failures really all that common?
Title: Re: SSD endurance test
Post by: 2bighorn on November 27, 2013, 09:03:07 PM
Many naysayers here are spreading FUD about SSD:s, especially the cheap TLC models.

In a heated argument about the claimed short liveness of the Samsung EVO series I dug up an endurance experiment that was an ongoing project as you can only write and erase so much data in so much time. Sadly it lacks the exact model but it does have a TLC based 840 on it.

So how has the experiment gone? As expected the TLC based Samsung died a horrible dead long time ago while... no wait! What's going on? It's still working and still as fast! The Intel 330 which was rated only for a 22Tb write life has survived for 300 Tb and has virtually no dead cells? Even the cheapest TLC is still running great after 300Tb.

http://techreport.com/review/25681/the-ssd-endurance-experiment-testing-data-retention-at-300tb


Bleh. Test with 6 SSDs LOL. Says nothing about longevity of SSDs because most of them die suddenly with no warning (controller failure). Cell wear is really only small part of the complex problem.

From one of the big data centers:
45% of 1st generation SSDs died after a year, only 5% made to 3rd year.
2nd generation is roughly 50% better, no data about 3rd gen yet (still ongoing).

As for the consumer market, SSD return rate was slightly higher than for the HDs, especially some brands (OCZ - no wonder they went tits up). That same data suggests Intel SSDs are still the best when it comes to reliability and longevity.

Title: Re: SSD endurance test
Post by: MrRiplEy[H] on November 27, 2013, 10:53:43 PM
Bleh. Test with 6 SSDs LOL. Says nothing about longevity of SSDs because most of them die suddenly with no warning (controller failure). Cell wear is really only small part of the complex problem.

From one of the big data centers:
45% of 1st generation SSDs died after a year, only 5% made to 3rd year.
2nd generation is roughly 50% better, no data about 3rd gen yet (still ongoing).

As for the consumer market, SSD return rate was slightly higher than for the HDs, especially some brands (OCZ - no wonder they went tits up). That same data suggests Intel SSDs are still the best when it comes to reliability and longevity.



The controller failures were due to bad firmwares which have been fixed a long time ago. If the return rate was only slightly higher than HDs considering all the firmware related deaths one can pretty safely say there are no endurance problems to speak of. Or reversely if there is, you have to raise hell about HDDs also, how come nobody does that? :)
Title: Re: SSD endurance test
Post by: 2bighorn on November 28, 2013, 01:23:56 AM
The controller failures were due to bad firmwares which have been fixed a long time ago.

That's what OCZ marketing guy would say. But the truth is, whilst some bugs were fixed, new ones emerged. Things are improving but not nearly as fast as SSD manufacturer are claiming.



If the return rate was only slightly higher than HDs considering all the firmware related deaths one can pretty safely say there are no endurance problems to speak of.

That's like talking about winning a marathon after running less than a mile. Premature. SSDs weren't around long enough.


Or reversely if there is, you have to raise hell about HDDs also, how come nobody does that? :)

Because SSDs were considered to be superior in any imaginable way but capacity. Why would anyone complain about old tech HDs?
Title: Re: SSD endurance test
Post by: MrRiplEy[H] on November 28, 2013, 03:30:34 AM
That's what OCZ marketing guy would say. But the truth is, whilst some bugs were fixed, new ones emerged. Things are improving but not nearly as fast as SSD manufacturer are claiming.

Source for your information, please.

Quote
That's like talking about winning a marathon after running less than a mile. Premature. SSDs weren't around long enough.

They have been here for 50 or so years. Implemented in super computers in the early 80's. That's ancient history in computer terms. Flash based SSDs came around 1995, again ancient history as far as computers go.

Quote
Because SSDs were considered to be superior in any imaginable way but capacity. Why would anyone complain about old tech HDs?

Or to rephrase that: Fear of unknown makes people cry about things which are not a problem in reality. Nobody is spewing doomsday scenarios about hdd:s.
Title: Re: SSD endurance test
Post by: Skuzzy on November 28, 2013, 05:56:16 AM
SSD's and HD's both suffer from one prevalent problem.  They can die on you, without warning.

I have dealt with customers who have had them fail.  They ran the Windows OS off of them, without moving the swap partition to another drive.  Worse, they fixed the size of the swap partition.  A good thing to do with a regular HD, but a bad thing to do with an SSD as it makes all those writes and deletes happen in a very small area of the drive.
Title: Re: SSD endurance test
Post by: MrRiplEy[H] on November 28, 2013, 08:04:21 AM
SSD's and HD's both suffer from one prevalent problem.  They can die on you, without warning.

I have dealt with customers who have had them fail.  They ran the Windows OS off of them, without moving the swap partition to another drive.  Worse, they fixed the size of the swap partition.  A good thing to do with a regular HD, but a bad thing to do with an SSD as it makes all those writes and deletes happen in a very small area of the drive.

All modern SSDs have had for a long time a wear leveling system where blocks are never reused twice in a row. If you append or refresh a file it gets written to an another set of memory cells. This way the ssds prevent failing from common tasks.
Title: Re: SSD endurance test
Post by: Delirium on November 28, 2013, 10:11:05 PM
Skuzzy, any plans to an option in AH to set film and write locations to a different drive than the installation drive/location?
Title: Re: SSD endurance test
Post by: Skuzzy on November 29, 2013, 07:01:28 AM
All modern SSDs have had for a long time a wear leveling system where blocks are never reused twice in a row. If you append or refresh a file it gets written to an another set of memory cells. This way the ssds prevent failing from common tasks.

The issue with with the "wear leveling systems" is once they start having errors, the ramp for potential hard failure is a very fast exponential curve, which almost goes vertical in a very short period of time.  It is not a perfect system and it can serve to extend the life of the SSD.  It usually does.

It does not prevent the original reasons for failure of the parts.  They do have a limited life.  It is the nature of the design.  It is a very high number of writes, but the limit is real.

I am not opposed to SSD's.  If I was going to use one, I would do everything I could to minimize writes to them.  I would never use an anti-virus program, in conjunction with one either.  One should expect a very long life, from one, when used in that manner.  Yes, it should exceed the life of a mechanical HD.

For me, I see no real reason to use one.  The speed at which an application loads is a pretty insignificant amount of time compared to using said applications.  Then again, I need a huge amount of storage space, which also makes SSD's useless for me.

Skuzzy, any plans to an option in AH to set film and write locations to a different drive than the installation drive/location?

Not at this time.
Title: Re: SSD endurance test
Post by: MrRiplEy[H] on November 29, 2013, 07:51:54 AM
The issue with with the "wear leveling systems" is once they start having errors, the ramp for potential hard failure is a very fast exponential curve, which almost goes vertical in a very short period of time.  It is not a perfect system and it can serve to extend the life of the SSD.  It usually does.

It does not prevent the original reasons for failure of the parts.  They do have a limited life.  It is the nature of the design.  It is a very high number of writes, but the limit is real.

I am not opposed to SSD's.  If I was going to use one, I would do everything I could to minimize writes to them.  I would never use an anti-virus program, in conjunction with one either.  One should expect a very long life, from one, when used in that manner.  Yes, it should exceed the life of a mechanical HD.

For me, I see no real reason to use one.  The speed at which an application loads is a pretty insignificant amount of time compared to using said applications.  Then again, I need a huge amount of storage space, which also makes SSD's useless for me.

You are obviously much more patient man than me. A snappy load time makes me happy every time, I never get tired of it.
Title: Re: SSD endurance test
Post by: 2bighorn on November 29, 2013, 03:57:51 PM
They have been here for 50 or so years. Implemented in super computers in the early 80's. That's ancient history in computer terms. Flash based SSDs came around 1995, again ancient history as far as computers go.

Specialty products, have almost nothing in common with current generation of enterprise/consumer SSDs.

Or to rephrase that: Fear of unknown makes people cry about things which are not a problem in reality. Nobody is spewing doomsday scenarios about hdd:s.

It's not fear of unknown, it's ABC of technology management.

Case in point (real world example, my LAN server)
(http://sierra-host.net/bb/456/8456/mountpoints.png)

Volume C is RAID1 with 2+1HS 1TB SAS HDDs
Volume E: is single 3TB SATA HD
Volume F: is RAID1 with 2 3TB SATA HDDs
Volume G: is RAID10 with 8+1HS 3TB SATA HDDs
Volume H: is RAID10 with 8+1HS 3TB SATA HDDs
They all hang behind cache pool with 2x180GB + 1x240GB SSDs

Lets say you replace all HDs with SSDs (forget for a moment all reliability concerns)
a) What are the performance gains if any?
b) Can you utilize performance gain without stepping up to, lets say 10GbE?
c) Cost? (probably high enough for me to sell the house and move into trailer in order to afford it)

Similarly, my main PC is used more or less as workstation, with RAID1 2x300GB SAS HDs and small SLC 20GB caching SSD. Since most of data get stored on the server, what's the point of increasing storage performance? I'm limited by network throughput.

It's really pointless to scream how good SSDs are. They have their place, but most often than not, performance gain is just not cost effective. Now, if you factor in all the longevity and reliability issues, they're just meh...

Anyway, there's a bright side too, you do get upgraded to SSD evangelist  :devil









 
Title: Re: SSD endurance test
Post by: MrRiplEy[H] on November 29, 2013, 04:46:15 PM
Specialty products, have almost nothing in common with current generation of enterprise/consumer SSDs.

It's not fear of unknown, it's ABC of technology management.

Case in point (real world example, my LAN server)
(http://sierra-host.net/bb/456/8456/mountpoints.png)

Volume C is RAID1 with 2+1HS 1TB SAS HDDs
Volume E: is single 3TB SATA HD
Volume F: is RAID1 with 2 3TB SATA HDDs
Volume G: is RAID10 with 8+1HS 3TB SATA HDDs
Volume H: is RAID10 with 8+1HS 3TB SATA HDDs
They all hang behind cache pool with 2x180GB + 1x240GB SSDs

Lets say you replace all HDs with SSDs (forget for a moment all reliability concerns)
a) What are the performance gains if any?
b) Can you utilize performance gain without stepping up to, lets say 10GbE?
c) Cost? (probably high enough for me to sell the house and move into trailer in order to afford it)

Similarly, my main PC is used more or less as workstation, with RAID1 2x300GB SAS HDs and small SLC 20GB caching SSD. Since most of data get stored on the server, what's the point of increasing storage performance? I'm limited by network throughput.

It's really pointless to scream how good SSDs are. They have their place, but most often than not, performance gain is just not cost effective. Now, if you factor in all the longevity and reliability issues, they're just meh...

Anyway, there's a bright side too, you do get upgraded to SSD evangelist  :devil

Nice troll. Yeah SSD:s have a "huge benefit" when they're used in a server behind a slow ethernet connection. Too bad we're talking about gaming pcs here and not your LAN server. Take any regular computer user from the street who uses a regular hdd for i/o and his transfer speeds will speed up 20 to hundred times simply by installing a 200 dollar SSD. About the price of your raid controller without the SAS drives lol. That makes night and day difference in performance.

It also seems weird that you saw the need for caching SSDs since you seem to think hdds are just fit for the bill.. ROFL!
Title: Re: SSD endurance test
Post by: 2bighorn on November 29, 2013, 05:58:28 PM
Nice troll. Yeah SSD:s have a "huge benefit" when they're used in a server behind a slow ethernet connection.

Intel I350T4 ie 4 Gbit/s trunk. That's pretty decent in my book. Most of us mortals can't afford 10Gbe

Too bad we're talking about gaming pcs here and not your LAN server.

Too bad majority of people use PCs for more than gaming.

Take any regular computer user from the street who uses a regular hdd for i/o and his transfer speeds will speed up 20 to hundred times simply by installing a 200 dollar SSD. About the price of your raid controller without the SAS drives lol. That makes night and day difference in performance.

Nobody disputes performance gains.

It also seems weird that you saw the need for caching SSDs since you seem to think hdds are just fit for the bill.. ROFL!

Why is that weird? I've never said I wouldn't use them, I've just said I would not use them as the only storage option. SSDs have way to go before any person who loves their data use them exclusively.
Title: Re: SSD endurance test
Post by: MrRiplEy[H] on November 30, 2013, 04:35:03 AM
Intel I350T4 ie 4 Gbit/s trunk. That's pretty decent in my book. Most of us mortals can't afford 10Gbe

Sorry to drop you down to earth but the majority of people still use 100mbit or 1Gbit lans if they even happen to have one. Most people own just a single computer and the ones that own more, typically network through wifi.

Quote
Too bad majority of people use PCs for more than gaming.

Yes too bad. How does it affect this discussion though? We're not talking about you here but the endurance of the SSDs!
Quote
Nobody disputes performance gains.

So what was the problem again? Nothing stops people from buying a relatively cheap 512Gb SSD and storing anything performance demanding (such as games and the OS) on it. Nothing stops people from having multiple 2Tb drives for video etc. storage on the side.

Quote
Why is that weird? I've never said I wouldn't use them, I've just said I would not use them as the only storage option. SSDs have way to go before any person who loves their data use them exclusively.

Who said they would be the only storage option? On a gaming machine though the SSD serves just fine as the only storage.
Title: Re: SSD endurance test
Post by: Skuzzy on November 30, 2013, 05:53:14 AM
You are obviously much more patient man than me. A snappy load time makes me happy every time, I never get tired of it.

In my system I do not see any cost/benefit to it.  Saving a second here and a second there, only when the executable is run the first time, does not seem to be the best investment I can make in my computer system.
Title: Re: SSD endurance test
Post by: MrRiplEy[H] on November 30, 2013, 05:59:58 AM
In my system I do not see any cost/benefit to it.  Saving a second here and a second there, only when the executable is run the first time, does not seem to be the best investment I can make in my computer system.

I guess it depends on what you do. If you frequently need to open files or different applications the load times are annoying. The quicker the system responds to the commands the better. That's why I love my puppy linux, can't beat running things straight from ram.

Title: Re: SSD endurance test
Post by: 2bighorn on November 30, 2013, 09:05:53 AM
Sorry to drop you down to earth but the majority of people still use 100mbit or 1Gbit lans if they even happen to have one. Most people own just a single computer and the ones that own more, typically network through wifi.

Yes too bad. How does it affect this discussion though? We're not talking about you here but the endurance of the SSDs!
So what was the problem again? Nothing stops people from buying a relatively cheap 512Gb SSD and storing anything performance demanding (such as games and the OS) on it. Nothing stops people from having multiple 2Tb drives for video etc. storage on the side.

Who said they would be the only storage option? On a gaming machine though the SSD serves just fine as the only storage.

Again, fact is that SSDs are (at minimum) 15 times more expensive than HDs ($/GB). This price difference is true whether you build a server with 1PB of storage or a PC with 1TB.
For that reason alone it would make more sense, even for a dedicated gaming machine, to have a hybrid storage system.

And then, there's that pesky reliability problem. Every time they think it's gone, they shrink cells and so we came from 100k cycles for SLC down to 1k cycles for TLC, which, according to you somehow increases endurance. Right...



Title: Re: SSD endurance test
Post by: Bizman on November 30, 2013, 11:46:50 AM
In my system I do not see any cost/benefit to it.  Saving a second here and a second there, only when the executable is run the first time, does not seem to be the best investment I can make in my computer system.
I have a feeling we are somewhat old fashioned, you and me... In my world, the longer something takes, the better I get paid. I charge per hour, even per minute if needed. All the time in the world is there for me to consume until the Pearly Gates call, and after that I just won't care any more. Of course there's deadlines every now and then, but they can't be helped whether my personal computer does things faster or not. On the other hand, I don't suffer from gastric ulcer or excessive stress, which seem to be the symptoms of a successful businessman...
Title: Re: SSD endurance test
Post by: MrRiplEy[H] on November 30, 2013, 11:57:01 AM
Again, fact is that SSDs are (at minimum) 15 times more expensive than HDs ($/GB). This price difference is true whether you build a server with 1PB of storage or a PC with 1TB.
For that reason alone it would make more sense, even for a dedicated gaming machine, to have a hybrid storage system.

And then, there's that pesky reliability problem. Every time they think it's gone, they shrink cells and so we came from 100k cycles for SLC down to 1k cycles for TLC, which, according to you somehow increases endurance. Right...





That 'pesky endurance problem' doesn't exist in real world as your favourite hdd:s have a shorter lifespan than modern SSD:s. But enough of that, time to leave this discussion.
Title: Re: SSD endurance test
Post by: Skuzzy on December 01, 2013, 06:39:45 AM
I guess it depends on what you do. If you frequently need to open files or different applications the load times are annoying. The quicker the system responds to the commands the better. That's why I love my puppy linux, can't beat running things straight from ram.

It depends on the application.  I deal with 40 to 60GB files all the time.  The applications are written for to deal with those files.  When I want to edit one, it is literally milliseconds in jumping to any point in the file.

It takes less than 2 seconds to open my editor(s), the first time.  The second time takes less than 1 second to open.  To open a large project (8+ 50GB video tracks, + 22+ sound tracks) takes about 15 seconds, the first time.  If I close and reopen that project, it takes seconds for it to open.  I can work for hours with instant responses.

It takes less than 9 seconds for my computer to boot to the desktop and be ready to use.

There are some very poorly written applications (games in particular) which have very poor loaders and probably could use the SSD.  I avoid applications like those simply for the fact, if they cannot get their loaders right, why would I trust them to get anything else right?  Personally, I detest masking problems, with poor code, by throwing more hardware at it.