Write speed always drops on Vertex 2 SSD

Phil2008 · 28 Mar 2011 at 14:16

So Ive been told after hrs of getting my head around it :eek:

I thought the write speed slow down was if you hammered the ssd with loads of writing, but I was wrong, as Ive secured erased(format) my ssd 3 times as the write speed kept dropping in about 3 weeks of secure erasing the ssd.

It was really bugging me at this point as I was scratching my head, why was the speed droping after every 3-4 weeks of starting a fresh.

So I went on the OCZ forum and said why does this keep happening. After meny posts later I kind of understand it. Basically After the ssd has written once to all af its chips or whatever ud call them, the write speed drops to about half, and duraclass now controls all writes and subsequent write. This is what causes the slowdown of the write speeds, and theres no way to stop this from happening or speeding the writes back up again, unless you secure erase it everytime the speed drops

But like this guy says on OCZ....... "you will see/feel the difference mostly only at benchmarks. To 85 percent you don't never reach a sequential write speed of more than 67MB/s at normal usage."

So for me once the write speed slows up,, compressible data is about 150mb/s and non compressible data is 67mb/s. But it varies from system to system.

I hope this makes sense to you guys as Im hopeless at explaining stuff :mad:

Phil2008 · 28 Mar 2011 at 16:48

You will always trigger the slow down of the write speed as windows is constantly writing bytes here and their. I trigger mine in 3-4 weeks of starting a fresh.

Isnt it slightly false advertising tho, saying writes upto 275mb/s, I mean yeah you can achieve that speed during the first few weeks but after that your lucky if you get 160mb/s?

Phil2008 · 28 Mar 2011 at 17:39

I mean thats why I couldn't understand why my write speeds kept slowing up, as no one tells you that the write speed is going to slow down after all the cells has been written to. I prob wouldn't of bought the drive if I knew this from the beginning :eek:

Phil2008 · 28 Mar 2011 at 20:56

DragonQ said:
Maybe your usage patterns are weird - Vertex 2s are best used as basic OS drives. Mine is as fast as the day it was new (5 months ago) with no secure erases since that first install and only 3 incompressible benchmarks.

You wont notice the speed loss until you benchmark the write speed

Heres mine:

A fresh install and about a month later.........

As this guy says..........

The SSDs with a Sandforce Controller seems to have 3 states.

The first is the fast one, all not used nand Flash has been cleared with a Secure Erase and the controller is able to write directly to the flash. But this mode ends, if every flash was used one time. On your drive, it could it be around 90GB of written data. You see the high speed by yourself with a write speed of 134MB/s.

The second one happens, if the “most” Flash was used to save any data ones.
You could have deleted the data again but the cells seem not to be cleaned up through the garbage collection to the same clean state as after a secure erase. To save new data to the "used" flash cells, the controller have first to clean it up (delete it) and only after this step, it is able to write the new data to the Flash cells. Could be your 64MB/s write speed.
On a few thread discussed, that a very long idle time help the drive to clean up the Flash cells at this use case to the better clean up state like after a “secure erase”. I have never tested it.

The third state is a mode to save the lifetime of the SSD and it slow down all the write speed of processes, which could be a benchmark or any other have load. It prevent a few benchmark user to kill all the flash erase cycles on the first big benchmark

and..............

2nd state is simply called "settled in state" and the controller has initiated a throttle to save life as well. Just not as severe of a throttle as the 3rd state. It's not that the controller must clean before write as it is simply just throttled by the firmware of the controller.

And yes, once all nand is hit, the drive will settle. Speed will never rise above this and the drive is designed to operate in this manner. Also consider that this drive is one of the top rated performance drives regardless of this incompressible write speed slowdown.

So what you see,.. is already factored into those ratings for this drive. Just use it as intended and if slowdowns/lags occur.. add more idle time to let GC keep up with your writing activity. Streaming video is the worst due to size and data type which hits more nand at the physical level.

Phil2008 · 28 Mar 2011 at 21:51

This is what their advertisement should have been for the read and write speeds like intels,,,,,,, BE TRUTHFULL

But the vertex 2 drives still are faster for writing compressible data tho, then the product on the link.

http://www.overclockers.co.uk/showproduct.php?prodid=HD-004-IN&groupid=701&catid=14&subcat=910

Phil2008 · 29 Mar 2011 at 12:27

You jammy thing you, I might try using perfect disk every month or so, maybe that will keep the write speeds up?

Phil2008 · 29 Mar 2011 at 12:46

Excellent thanks. Wonder if this idea will work, thanks to Vimes.

Phil2008 · 29 Mar 2011 at 13:14

Vimes said:
I really do wish that I had done a before and after benchmark when I ran Perfect Disk 11. It does show in its logs...

Number of Fragmented Files Before: 12215
Number of Fragmented Files After: 102
Number of Fragmented Directories: 541
Number of Excess Fragments: 31103
Number of Excess Fragments After: 229
Percent Fragmented Before: 9.4%
Percent Fragmented After: 0.1%

Recommend Defrag Only because of File Fragmentation is high

When it notices that the drive is an SSD it will not offer to defrag it but to consolidate the space. This I did.

Having done that in the past (this was the first time since November(ish) I have noticed that I maintain a consistently high benchmark result. Whether PD has an effect or it is just snake oil I'm not sure but there is a thread here you can read...

http://www.ocztechnologyforum.com/forum/showthread.php?78276-PERFECTDISK-AND-SSD-quot-S

..even though it is shown in that thread for our types of drives you should benchmark using ATTO, and not that often.!

Perfect Disk and SSD's...

http://www.raxco.com/support/kb/791

Im running perfect disk now, how often do you run it?

Phil2008 · 29 Mar 2011 at 13:19

It takes a while doesn't it,, I was expecting it to be done in a few secs?

Phil2008 · 29 Mar 2011 at 13:34

Ok, its done, I was kind of pooping myself while that was happening cos it looked like it was defraging in some areas, but I just run crystaldisk and everything is still performing ok. Guess we will just have to wait and see now and cross fingers and toes.

Phil2008 · 29 Mar 2011 at 13:53

I would have thought the file fragmentation would been still at 2% but its at 0% and the numbers before and after,,, So Im a bit worried there? :eek:

Phil2008 · 29 Mar 2011 at 15:07

Anyway thanks Vimes, I will see if it solves my prob and run PD once a month. I dont use ATTO as it writes stupid amounts of data, and Ive got crystal disk to write only 100mb, but its not the true reading as crystal disk uses uncompressible data. I also never ever log off

Maybe this thread will be useful for other vertex 2 users and I'll report back in about a month or 2 to let you know if its worked or not.

Phil2008 · 29 Mar 2011 at 15:24

Ive told these guys at OCZ that you use PD and havent had any probs with write speed, and ask then if this would solve my prob.. and their reply was............

highly unlikely as the drive is designed to throttle down through the Durawrite process. I don't believe for one second that PD11 nor any other software would cicumvent this internal process short of secure erasure. If his drive has not throttled yet, then he has not filled it completely. There are only a couple of guys around here that still subscribe to the "the drive may never throttle" theory. Pretty obvious, I'm not one of them.

And as for the ability of PD11 to reduce throttling? Don't see how that would be possible since all it does is to increase the OS/logical efficiency(with slight chance of some internal trimming/remapping/consolidation at the physical level) and additional writes are obviously occuring with its usage. The writes are usually minimal but there nonetheless. And the Windows consolidation has no where near the efficiency of PD11 consolidation as I've tested them side by side. IMO, PD11 is best served to optimize images slated for backup restoration. That way everytime you flash the image over after an SE, it's nice and efficient from the start.
..........................................................

But we will see...........

Phil2008 · 29 Mar 2011 at 17:52

Vimes do you think you can run Crystaldisk plz and use the same settings as im my screen shot. If the throttling has kicked in, it will be in the double figures and if throttling hasn't kicked in, it should be 130mb upwards. As OCZ say that even if throttling has kicked in ATTO wont be much diff, but I find that hard to believe.

Heres my ATTO atm

Phil2008 · 29 Mar 2011 at 20:10

Vimes said:
Here is mine...

mine seems to be "throttled". Albeit that is the very first time I have run that benchmark as I have always been under the impression that the ATTO is more applicable for our types of drives.?

Still your ATTO results are similar to mine.

I'm not sure though what are you suggesting is the implication or problem with what is happening, as in everyday use it seems fine..?
Unless you are just learning about the differences in advertising "up to" and then real life "actuals"

Ummmmmmm, back to square 1. Thread closed :eek:

Phil2008 · 29 Mar 2011 at 20:38

Vimes said:
lol

I give up trying to work out these ssd's

Phil2008 · 30 Mar 2011 at 21:07

Before I ran PD, the disk map thing was in a right mess when I checked it in defraggler, look at it now..... Some bits has moved about now tho.

Phil2008 · 5 Apr 2011 at 11:35

Just dropped to 67mb in crystal disk again :mad:

but ATTO is still nice and high.

Phil2008 · 5 Apr 2011 at 12:15

Phil2008 said:

so this the speed for all uncompressible data(crystaldisk) and the speed for all compressible data(ATTO)?