Do all SSDs Throttle Writes after a Period of Continuous Writes?

[squick3n]

Hello AnandTech Forums

So as I understand it, pounding away on a Sandforce drive with lots of writes will trigger a sort of defense mechanism and the drive will progressively slow down until it is given time to recover. Not just the "backed into a corner" test, but such that even after a secure erase, it will still write slowly.

Is this just a "feature" of the SF drives, or do all SSDs do this? I had been under the impression that a secure erase to a drive would restore it to a "fresh" state, but apparently that is not the case, at least with SF.

Thanks for any help!

 

[DirkGently1]

No. The Intel 320 hasn't slowed down at all in these tests. 135TB written, MWI 30%. (40GB drive, 135TB data written = writing over the entire drive 3375 times)

http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm

The Sandforce didn't fare well though. Unrecoverable throttle.

 

[squick3n]

Thanks for the link. It's a very strange feature of the SF drives Why not just let you do what you want with your drive? I don't think anyone warranties used up NAND, so what's the point?

Is that the only test out there? Intel controller seems to avoid it, but what about Marvell?

 

[tommylommykins]

My understanding was that this is a feature that's there explicitly to cripple drives...

The idea being that later on, SSDs can be sold without this lock for a premium, because they have higher performance

 

[Emulex]

the reserve space is used differently with sandforce. try to get a spec sheet on the SF-12xx and it will explain how it uses reserve space to boost performance. think of it as pre-provisioning - or like intel does turbo-boost - it's not mean to work that way 100% of the time. if you hacked the bios to work that way 100% of the time you'd end up with a dead drive in 6 months and their return rates would be too high and they'd get dropped from the channel.

 

[groberts101]

I believe I may have passed the 100 mark for SE's on most of my Sandforce drives(I do it as a measure of protection to avoid settled in states(some call that a "lifetime throttle". If the ATA secure erase command is passed properly to the drives firmware(all drives have that written/added to the code), then it will return fresh speeds EVERYTIME. You cannot reset smart data(only destructive engineering based flashes can do that) but it will FOR SURE reset non-GC mapped fresh speeds. read here to understand more.

http://www.ocztechnologyforum.com/f...ives-TRIM-OP-area-use-and-Life-write-throttle
So again.. if the drive is not wiping the GC/Durawrite maps?.. then the SE command is not being initiated properly.

 

[squick3n]

I'm confused then, b/c the posts on xtremesystem (on the durability test where the guy's V2 would only write at 5 mb/s) and on ocz's forums indicate that a SE cannot undue a throttled state

http://www.ocztechnologyforum.com/forum/showthread.php?90401-Vertex-3-Lifetime-Write-Throttle

 

[Mark R]

I'm confused then, b/c the posts on xtremesystem (on the durability test where the guy's V2 would only write at 5 mb/s) and on ocz's forums indicate that a SE cannot undue a throttled state

http://www.ocztechnologyforum.com/forum/showthread.php?90401-Vertex-3-Lifetime-Write-Throttle

Unfortunately, no one official is saying anything about this particular case. However, it appears to be that there is an 'abuse' throttle, that is triggered when the drive detects deliberate attempts to destroy the drive (by constantly writing such huge amounts of data that you would expect to kill the drive within a few months).

The guy in the durability thread who did get his drive into this state did so by continuously running random writes at 30 MB/s for 2 weeks (that's a load 100 times heavier than a typical user).

 

[squick3n]

Unfortunately, no one official is saying anything about this particular case. However, it appears to be that there is an 'abuse' throttle, that is triggered when the drive detects deliberate attempts to destroy the drive (by constantly writing such huge amounts of data that you would expect to kill the drive within a few months).

The guy in the durability thread who did get his drive into this state did so by continuously running random writes at 30 MB/s for 2 weeks (that's a load 100 times heavier than a typical user).

But I guess I can infer that this write throttling is limited only to Sandforce. A secure erase of C300/M4 or 510/320 drives will bring them back to fresh, even if you've been writing heavily.

 

[Mark R]

But I guess I can infer that this write throttling is limited only to Sandforce. A secure erase of C300/M4 or 510/320 drives will bring them back to fresh, even if you've been writing heavily.

This does seem to be a Sandforce only issue. Those people doing endurance tests on the Intel drives don't seem to have hit a 'lifetime throttle' limit, even though early reports about the Intel drives suggested that one existed.

 

[groberts101]

you are interpreting things the way you see them based on one particular user who was not able to properly pass the ATA secure erase command to the controllers firmware or simply has a faulty drive. The situation in that durability test is not the norm and I myself have NEVER EVER seen someone who was not able to SE out of all throttled state with these Sandforce controllers. Tony himself even told you guys that there is a lifetime throttle in that OCZ thread linked above.

The confusion comes from the fact that it's called a lifetime throttle when in fact the SE command will clear the Durawrite's GC map and release any and all throttles. I do it purposely to get rid of lifetime throttles which are actually better said to be "settled state throttles". Settled state is there for life and no amount of idle time can recover back to fresh incompressible write speed. Hence.. the term "lifetime throttle". However.. a secure erase clears the Durawrite mapping that's required to implement it. No map?.. no throttle.

I also know for fact because I have helped and seen hundreds upon hundreds of folks on the OCZ forums who have put their Sandforce drives into what's called a "hammered state" throttle. It's common knowledge that SE will release the 2 distinct stages of throttling with Sandforce controlled drives. Settled state AND hammered state. Idle time will only recover UP TO a settled state speed when the drive is coming out of a hammered state.

Go ask over there if you don't believe me. Be specific with your question as to not confuse yourself and others even more than you already are.

Ask.." does a secure erase return factory fresh speeds(as in non-lifetime throttled) on Sandforce controlled drives?".. or maybe.. does secure erase release the "lifetime throttle" on a Sandforce drive?".. and you will get the correct answer.

 

[squick3n]

http://www.ocztechnologyforum.com/f...se-on-an-Vertex-3-Release-a-Lifetime-Throttle

done

 

[groberts101]

hate to say I told ya so.. BUT.. I will. LOL

Only teasin' bud.. my pride is not that sensitve to being wrong, but I've enough experience with SE(along with seeing/helping literally hundreds) that I can easily make factual blanket statements without worry about it being bruised.

PS. it should be remembered that the original sata2 drives have much larger throttling tendencies than the newer gens. VERY hard to keep a SF2281 controller down in a hammered state for more than a little bit. The recycling engine and on-the-fly maintenance is much improved over the first iterations.

 

[squick3n]

I didn't ask b/c I doubted anyone. There was conflicting information, and I was glad to get OCZ's input on it. Looking at the updated tests from xtremesystems, even with the SF-1200, he had to write an unreasonable amount to hammer it down. But he also idled it for a week, which is a bit unreasonable.

 

[groberts101]

yeah.. it takes a bit of larger incompressible/random data streams to hammer them. Is based on capacity of course and the poster over there has the slowest and smallest sandforce drive of all.

The bigger issue there is the fact that he was unable to clear the drives maps and start fresh back to factory new speeds. This was further evidenced by the fact that only shortly after tyhe attempted SE.. he was right back to a heavily throttled state.

IMHO, he was not being hampered by a throttle so much as another underlying transfer issue or problem with the controller itself. I know this because even the worse case throttling of a sandforce controller is still capable of writing sequentially at 50% of fresh speed when using incompressible data. Another Sandforce fact there.

The only ohter guess I would have is that the physically free block structure of that drive was so heavily fragmented by the random data flow he was barraging that tiny little half channeled drive with.. was creating havoc for its ability to read-write-modify the additional data flows. Some who have tried to use those smaller drives in servers have found out the hard way that small capacity Sandforce's on the fly cleaning is not up to the task. They however were always able to SE their way back to fresh speeds. Until, the random data flow was too much again, that is.

 

[Mr.Peepers]

The bigger issue there is the fact that he was unable to clear the drives maps...

IMHO, he was not being hampered by a throttle so much as another underlying transfer issue or problem with the controller itself. I know this because even the worse case throttling of a sandforce controller is still capable of writing sequentially at 50% of fresh speed when using incompressible data. Another Sandforce fact there.

Care to correct your erroneous "factual" statements now that you know you made some bad assumptions?

Does Life-Time Throttle ring any bells?

http://www.xtremesystems.org/forums...e-Throttling&p=4891332&viewfull=1#post4891332

 

[groberts101]

LOL.. sad thing is that those guys think I'm full of it because of one bad assumption on my part. There are so many other bad assumptions being made over there about Sandforce it's not even funny. The fact is that those statements in "general terms" are still very much valid.

So in the end?.. I harldy doubt me being proven wrong about one Sandforce fact cancels out all the other misinformation and assumptions made by others.

Not to mention aside from that extreme endurance testing?.. who the hell would ever see it anyways? Is moot'er than moot for anyone not writing 35 Terabytes within a short timeframe. lol

I've written over 1 Terabyte in less than 4 hours to my drive and SE restores full fresh speeds without issue. Good enough for me, I guess.

I was wrong. Big whoop.. what'ya want me to say besides that? Sorry for hurting "sensitive nerd" feelings? hehe

So here it is for all those who are so into rubbing noses in **** around here.

Sandforce does have "Lifetime throttling" written into the code. Write numerous Terabyte's worth of data in a very short timeframe?.. then you get "Lifetime Throttled" until Sandforce's code implemented by the mfgr for that particular drive(which in that case was only the smallest slowest of them all to be sure) decides that the data flow over a given amount of time has decreased enough to let off the throttle. SE will NOT release this. "Yayyy... they were right and that coc*y OCZ basta** was wrong!" LOL

Then from there we have the "real life" throttles that only typically used scenarios would ever see. Such as the "Settled State Throttle" which is considered the normal day to day running speed once all nand has been written and the Durawrite/GC map has been fully formed. Yep.. you can SE your way right out of this.

Then we have the "power user", "benchmark fanatic", and "I don't care if it's not SLC.. I want to use it in my server anyways" type of throttle called a "Hammered State". Yep.. you guessed it.. you can SE you way back to fresh from this state as well.

So not much more than ego's are going to be petted up here since no one would ever hit that type of throttle. Especially since most don't use a small 40GB bottom of the Sandforce tier drive like that to try and write 35 Terabytes worth of data. So your point?.. is rather pointless aside from some petty satisfaction gained from it.

Guess it also proves I'm not as extreme as I thought I was with my drives and I'll have to beat them even harder next time I test them.