Long term performance of SSDs. Analysis of Intel SSDs.

[taltamir]

Corsair S128 is selling for 327$ on newegg. It has 128GB of storage, 110MBps read and 80MBps write.

Intel X25-M is selling for 369$ on newegg. It has 80GB of storage, 250MBps read, and 80MBps write.

Regular people, with slower desktop and laptop who only have SATAI interface, should they spend 40$ more on Intel that is also 40GB smaller and doesn't really outperform that much when limited by SATA I 120MBps real life limit?

The SATA1 interface has 1.5gbps bandwidth. 1500/8 = 187.5MB/s
However it has 2 extra bits of overhead per 8 bits transfered, so you divide it out by 10 not by 8. So 1500 / 10 = 150MB/s.
I don't know where you pulled a 120MB/s limit out of.

Furthermore, the intel drives suffer from steadystate issues, WHICH THIS THREAD IS ABOUT BTW!!!!!, and the Samsung drives DON'T!

You couldn't be more wrong. Steady state means that the drive has had all sectors written too, which causes fragmentation of free space, when you try to write something the drive has to read the entire contents of the area it is going to erase, then erase it, then write back the content (new + old). Defragging with traditional defraggers will make matters worse due to address locations not matching the physical locations (it is constantly remapped to perform wear leveling). If you have a drive without wear leveling it will still suffer from it, but to a lesser degree and it will be reduceable with defragging (BUT ONLY IF IT HAS NO WEAR LEVELING - which means the drive will not last long)

As there is no single SSD without wear leveling, it is an issue that affects ALL drives without exception

However the intel logic controller has optimizations that reduce the amount steadystate affects, at least according to what I have read. Furthermore, it has NCQ, which has been TESTED by independant reviewers to greatly decrease the steady state penalty.

So in fact, the samsung drive is not a magic SSD that has no steady state performance issue, it actually has a WORSE steadystate problem then the intel drive.

I have to say that I am getting really sick and tired of all these SPAM attempts by SSD mfctrs on forums!

Me too, so how about you stop pushing the inferior samsung product. Anandtech itself is the major source of info on the subject and the superiority of the intel SSD. You however seem to be talking out of your butt... AND accuse me of being a shill for intel.

 

[taltamir]

PS. the nature of steadystate is such that is not something you can ship a drive in, it not even a constant, it varies on a file by file status. The best way to describe it is as the unholy bastard child of the raid5 write hole and fragmentation.

 

[fuzzybabybunny]

Originally posted by: Viper GTS
Our use pattern is extremely unusual. It's multiple streams of non-stop writes, of varying sizes, and heavily randomized. This could potentially be a 24/7 use. To make matters worse we need at a minimum 1 TB capacity. We currently use 8x450 GB 15K SAS, & for a couple very specific uses where reliability/ruggedness is more important than capacity we're willing to take the capacity hit to 1 TB.

I would not hesitate to put one of these drives in a laptop. In fact if I were buying a laptop today I would insist on it. They're perfect for that. For my personal desktop use I use Velociraptors, and will continue to use them for the foreseeable future.

Viper GTS

What would you recommend for my situation in the OP? The Runcore drives I'm waiting on AREN'T like these 150+/100+ MB/s SSD drives that everyone else can put in their lappies. I'm confined to 1.8" ZIF drives, and these SSDs only do 80/45 compared to the 25MB/s that I'm currently getting with my rotary drive. Should I just cancel the order because after the SSD reaches steady state it isn't going to be that much better than even my rotary drive?

 

[Viper GTS]

Originally posted by: fuzzybabybunny

What would you recommend for my situation in the OP? The Runcore drives I'm waiting on AREN'T like these 150+/100+ MB/s SSD drives that everyone else can put in their lappies. I'm confined to 1.8" ZIF drives, and these SSDs only do 80/45 compared to the 25MB/s that I'm currently getting with my rotary drive. Should I just cancel the order because after the SSD reaches steady state it isn't going to be that much better than even my rotary drive?

I would probably go for the SSD.

Most of what I'm posting about is under extremely abusive write patterns. For laptops I would have no hesitation about SSD's, & would insist on it if it were my own.

Viper GTS

 

[MikhailT]

Originally posted by: taltamir

Corsair S128 is selling for 327$ on newegg. It has 128GB of storage, 110MBps read and 80MBps write.

Intel X25-M is selling for 369$ on newegg. It has 80GB of storage, 250MBps read, and 80MBps write.

Regular people, with slower desktop and laptop who only have SATAI interface, should they spend 40$ more on Intel that is also 40GB smaller and doesn't really outperform that much when limited by SATA I 120MBps real life limit?

The SATA1 interface has 1.5gbps bandwidth. 1500/8 = 187.5MB/s
However it has 2 extra bits of overhead per 8 bits transfered, so you divide it out by 10 not by 8. So 1500 / 10 = 150MB/s.
I don't know where you pulled a 120MB/s limit out of.

Yes, that is the max transfer rate in theory, nothing ever close to the limit of 150MBps. I just took off 30MBps for overhead from controllers, cables, mobo, etc etc and the fact i seen many hard drives hitting less than 120MBps even SSD that is well over 200MBps couldn't get over 120MBps with SATA I interface.

 

[coolVariable]

Originally posted by: taltamir
PS. the nature of steadystate is such that is not something you can ship a drive in, it not even a constant, it varies on a file by file status. The best way to describe it is as the unholy bastard child of the raid5 write hole and fragmentation.

Are you just disagreeing with everything on purpose, because you don't seem to know what you are talking about.

Samsung and Mtron ship their SSDs after a full write cycle, where every cell has been written to at least once. At that point, in order to write to any cell, the contents need to be read, the cell information erased and the new information written. This is the so-called steady-state. That's why the Samsung and Mtron drvies have no (or only very little) performance fluctuations. OCZ, intel, etc.don't do this which is why you see the performance drop off mentioned in the article linked in the first post of this thread.

 

[Zap]

Originally posted by: fuzzybabybunny
What would you recommend for my situation in the OP? ... put in their lappies.

It's already been mentioned by a few that SSDs can be a great solution for notebook use. It just isn't the be-all end-all (yet?) for the overall long-term performance crown.

 

[garikfox]

If you buy a recent SSD now, It will last Longer then a HD.

i.e - If you write too it 50gb a day it will last, 187 years. LOL

 

[IntelUser2000]

Originally posted by: coolVariable

Originally posted by: taltamir
PS. the nature of steadystate is such that is not something you can ship a drive in, it not even a constant, it varies on a file by file status. The best way to describe it is as the unholy bastard child of the raid5 write hole and fragmentation.

Are you just disagreeing with everything on purpose, because you don't seem to know what you are talking about.

Samsung and Mtron ship their SSDs after a full write cycle, where every cell has been written to at least once. At that point, in order to write to any cell, the contents need to be read, the cell information erased and the new information written. This is the so-called steady-state. That's why the Samsung and Mtron drvies have no (or only very little) performance fluctuations. OCZ, intel, etc.don't do this which is why you see the performance drop off mentioned in the article linked in the first post of this thread.

The drop is there. The old SLC drives use mediocre write levelling and the performance degradation is less. The new MLC Samsung drives with newer controller does suffer from it. PCWatch had a review about this.

 

[coolVariable]

Originally posted by: IntelUser2000

Originally posted by: coolVariable

Originally posted by: taltamir
PS. the nature of steadystate is such that is not something you can ship a drive in, it not even a constant, it varies on a file by file status. The best way to describe it is as the unholy bastard child of the raid5 write hole and fragmentation.

Are you just disagreeing with everything on purpose, because you don't seem to know what you are talking about.

Samsung and Mtron ship their SSDs after a full write cycle, where every cell has been written to at least once. At that point, in order to write to any cell, the contents need to be read, the cell information erased and the new information written. This is the so-called steady-state. That's why the Samsung and Mtron drvies have no (or only very little) performance fluctuations. OCZ, intel, etc.don't do this which is why you see the performance drop off mentioned in the article linked in the first post of this thread.

The drop is there. The old SLC drives use mediocre write levelling and the performance degradation is less. The new MLC Samsung drives with newer controller does suffer from it. PCWatch had a review about this.

Link?
Samsung and Mtron ship their SSDs after a full write cycle, that is steady-state!
Don't really see where you could get another performance drop from.

 

[Idontcare]

Originally posted by: coolVariable
At that point, in order to write to any cell, the contents need to be read, the cell information erased and the new information written. This is the so-called steady-state.

Is there some reason the drive waits to prepare the cells for a write?

It would seem like the thing to do here is actively prepare (in the background of course, when the drive is otherwise idle) previously used cells for a new write so that when that time comes there is as little delay in writing to the cells as possible.

 

[taltamir]

Link?
Samsung and Mtron ship their SSDs after a full write cycle, that is steady-state!
Don't really see where you could get another performance drop from.

Then you obviously don't understand what steady state is.

 

[taltamir]

Originally posted by: Idontcare

Originally posted by: coolVariable
At that point, in order to write to any cell, the contents need to be read, the cell information erased and the new information written. This is the so-called steady-state.

Is there some reason the drive waits to prepare the cells for a write?

It would seem like the thing to do here is actively prepare (in the background of course, when the drive is otherwise idle) previously used cells for a new write so that when that time comes there is as little delay in writing to the cells as possible.

that would be doing the OPPOSITE of wear leveling in terms of wasting write cycles and lifespan of the drive.

 

[Idontcare]

Originally posted by: taltamir

Originally posted by: Idontcare

Originally posted by: coolVariable
At that point, in order to write to any cell, the contents need to be read, the cell information erased and the new information written. This is the so-called steady-state.

Is there some reason the drive waits to prepare the cells for a write?

It would seem like the thing to do here is actively prepare (in the background of course, when the drive is otherwise idle) previously used cells for a new write so that when that time comes there is as little delay in writing to the cells as possible.

that would be doing the OPPOSITE of wear leveling in terms of wasting write cycles and lifespan of the drive.

How so? The cells were going to be written to eventually, so preparing them for that eventuality in advance isn't increasing the number of times the cells are written to.

 

[Viper GTS]

Originally posted by: coolVariable
Link?
Samsung and Mtron ship their SSDs after a full write cycle, that is steady-state!
Don't really see where you could get another performance drop from.

There is more to it than just a lack of unwritten cells available. Intel documents that it varies by workload/access pattern, and switching workloads requires a new cycle of conditioning the drive to obtain steady-state results.

Originally posted by: Idontcare

How so? The cells were going to be written to eventually, so preparing them for that eventuality in advance isn't increasing the number of times the cells are written to.

Because you would be burning a write cycle by "preparing" them and again when you actually populate it with data. Great for performance, horrible for write amplification.

Viper GTS

 

[Idontcare]

Originally posted by: Viper GTS

Originally posted by: Idontcare

How so? The cells were going to be written to eventually, so preparing them for that eventuality in advance isn't increasing the number of times the cells are written to.

Because you would be burning a write cycle by "preparing" them and again when you actually populate it with data. Great for performance, horrible for write amplification.

Viper GTS

You (and talta) are right, I wasn't thinking about that. The trade-off would be an increase in write amplification.

 

[coolVariable]

Originally posted by: Idontcare

Originally posted by: Viper GTS

Originally posted by: Idontcare

How so? The cells were going to be written to eventually, so preparing them for that eventuality in advance isn't increasing the number of times the cells are written to.

Because you would be burning a write cycle by "preparing" them and again when you actually populate it with data. Great for performance, horrible for write amplification.

Viper GTS

You (and talta) are right, I wasn't thinking about that. The trade-off would be an increase in write amplification.

No because write amplification is something else:
http://www.extremetech.com/art.../0,2845,2329594,00.asp

 

[coolVariable]

Originally posted by: taltamir
Then you obviously don't understand what steady state is.

My friend, I suggest you do some research before posting on this forum again.

 

[taltamir]

that website you linked coolvariables, poorly represents the nature of write amplification, the anandtech articles explain it better. And write amplification is EXACTLY what we used it to describe.

The problem is that SSD cannot ERASE individual blocks, they must bulk erase. they might be able to write in 4K blocks, but only erase in 128K groups at once (the extremetech article uses that example, but anandtech said 512KB needs to be erased at once IIRC). That means that to write 4kb you have to read the other 124kb that reside on the erasing CIRCUIT that that 4k block is on... Even if it DOES have free space on it, the controller cannot TELL because it is not aware of the FILESYSTEM in use. You would need special filesystems that inform the controller of not using a block. Moreover the ideal for speed would be to do background "defragging" of those erase circuits... but that would be even worse than having no wear leveling at ALL.
Intel claims to have done a lot of work to improve this situation, as well as having NCQ which allows them to combine writes for a similar erase circuit. (while a spindle drive wants to organize writes to be in order to prevent wasted time spinning, an SSD want to corrolate all the writes going to one erase circuit and do only one read-erase-write cycle.

This, btw, is reminiscent to the write performance issues of raid5:

RAID 5 implementations suffer from poor performance when faced with a workload which includes many writes which are smaller than the capacity of a single stripe;[citation needed] this is because parity must be updated on each write, requiring read-modify-write sequences for both the data block and the parity block. More complex implementations may include a non-volatile write back cache to reduce the performance impact of incremental parity updates.
Random write performance is poor, especially at high concurrency levels common in large multi-user databases. The read-modify-write cycle requirement of RAID 5's parity implementation penalizes random writes by as much as an order of magnitude compared to RAID 0.

 

[IntelUser2000]

Originally posted by: coolVariable

Originally posted by: IntelUser2000

Originally posted by: coolVariable

Originally posted by: taltamir
PS. the nature of steadystate is such that is not something you can ship a drive in, it not even a constant, it varies on a file by file status. The best way to describe it is as the unholy bastard child of the raid5 write hole and fragmentation.

Are you just disagreeing with everything on purpose, because you don't seem to know what you are talking about.

Samsung and Mtron ship their SSDs after a full write cycle, where every cell has been written to at least once. At that point, in order to write to any cell, the contents need to be read, the cell information erased and the new information written. This is the so-called steady-state. That's why the Samsung and Mtron drvies have no (or only very little) performance fluctuations. OCZ, intel, etc.don't do this which is why you see the performance drop off mentioned in the article linked in the first post of this thread.

The drop is there. The old SLC drives use mediocre write levelling and the performance degradation is less. The new MLC Samsung drives with newer controller does suffer from it. PCWatch had a review about this.

Link?
Samsung and Mtron ship their SSDs after a full write cycle, that is steady-state!
Don't really see where you could get another performance drop from.

I did say PCWatch did I not?

http://pc.watch.impress.co.jp/...9/0212/hirasawa014.htm

And taltamir, I'm pretty sure its not NCQ that does the wear levelling part of the Intel SSD because what if NCQ was disabled?? The internal controller uses write combining algorithms to do that.

Of course, you are right about everything else. I'm thinking one another reason they don't do background cleaning is because to keep idle power usage low.

 

[taltamir]

And taltamir, I'm pretty sure its not NCQ that does the wear levelling part of the Intel SSD because what if NCQ was disabled?

I never said NCQ does wear leveling, i said NCQ aware wear leveling can reduce the impact of steady state and your write amplification. And according to one of the articles on the subject the intel controller does just that... but I have only read it in one location and it might be bull. However it makes sense.
Wear leveling at its simplest will try to, by itself, write each block to the least written to block on the drive. If you combined it with NCQ it will still try to do the same, but it will now have the added benefits of doing read (512K circuit)-modify with SEVERAL pending 4k writes at once-erase entire 512K-write modified data.

 

[coolVariable]

Originally posted by: IntelUser2000

Originally posted by: coolVariable

Originally posted by: IntelUser2000

Originally posted by: coolVariable

Originally posted by: taltamir
PS. the nature of steadystate is such that is not something you can ship a drive in, it not even a constant, it varies on a file by file status. The best way to describe it is as the unholy bastard child of the raid5 write hole and fragmentation.

Are you just disagreeing with everything on purpose, because you don't seem to know what you are talking about.

Samsung and Mtron ship their SSDs after a full write cycle, where every cell has been written to at least once. At that point, in order to write to any cell, the contents need to be read, the cell information erased and the new information written. This is the so-called steady-state. That's why the Samsung and Mtron drvies have no (or only very little) performance fluctuations. OCZ, intel, etc.don't do this which is why you see the performance drop off mentioned in the article linked in the first post of this thread.

The drop is there. The old SLC drives use mediocre write levelling and the performance degradation is less. The new MLC Samsung drives with newer controller does suffer from it. PCWatch had a review about this.

Link?
Samsung and Mtron ship their SSDs after a full write cycle, that is steady-state!
Don't really see where you could get another performance drop from.

I did say PCWatch did I not?

http://pc.watch.impress.co.jp/...9/0212/hirasawa014.htm

And taltamir, I'm pretty sure its not NCQ that does the wear levelling part of the Intel SSD because what if NCQ was disabled?? The internal controller uses write combining algorithms to do that.

Of course, you are right about everything else. I'm thinking one another reason they don't do background cleaning is because to keep idle power usage low.

[sarcasm]
Now that article is an awesome read!
Though these articles say the exact opposite:
http://news.cnet.com/8301-1392...news&tag=2547-1_3-0-20
http://www.tecchannel.de/stora...enchmark_transferrate/
[/sarcasm]


I have to say, you and taltamir really take the crown for having NO CLUE but acting like you're experts. Are you completely full of it or just 99%?

 

[taltamir]

if you think my explanation of how and why steadystate is an issue is wrong, please explain it correctly.

 

[coolVariable]

Originally posted by: taltamir
if you think my explanation of how and why steadystate is an issue is wrong, please explain it correctly.

I did:

Originally posted by: coolVariable
Samsung and Mtron ship their SSDs after a full write cycle, where every cell has been written to at least once. At that point, in order to write to any cell, the contents need to be read, the cell information erased and the new information written. This is the so-called steady-state. That's why the Samsung and Mtron drvies have no (or only very little) performance fluctuations.

 

[taltamir]

this in no way shape or form contradicts my explanation of what steady state is.