More important than Benchmarks

[Nec_V20]

All the time I see reports of this or that SSD drive and its benchmark performance, the one test that is never performed is to see if the drive actually works throughout its specified capacity.

 

There is a really nice small utility for doing this and it is called "H2testw" it is a German program but the language can be switched to English.

 

You can get it here:

 

It was created for and by one of the most respected computer magazines in the World and one which I have been reading for over 25 years now called "c't magazin für computertechnik". The link I provided is to the English language site of that magazine.

 

It does not need to be installed it runs standalone.

 

What this utility does is write to the SSD until it is full and then tests if what has been written can actually be read. The results, if negative, cannot be argued away as some aberration of some dodgy software because there is a highly respected authority behind this utility. And because the download link I provided is to the magazine you know that it has not been tampered with.

 

Here is a test result from my new Neutron 120 GB SSD. I had done a complete run on it before I built it into my system and I have done another one as an example of the result on just 10 GB to show you a sample of the result:

 

Warning: Only 10000 of 114370 MByte tested.

Test finished without errors.

You can now delete the test files *.h2w or verify them again.

Writing speed: 158 MByte/s

Reading speed: 169 MByte/s

H2testw v1.4

 

I ran the test through twice on the entire SSD and both times it came out without errors before I installed it into my new system.

 

An SSD is not exactly cheap and you also don't want to find out any defects it might have after you have started to use it. You want to know as soon as you have it out of the box whether or not it is in fully working order.

 

The good thing is that if the utility comes up with errors you can print off the result, burn the result to a CD and include the utility so that the manufacturer can run the test as well, with regard to an RMA.

[canthearu]

You shouldn't be running this test on anything other than USB keys that you feel may be fake. SSDs bought on the legitimate market will never show a fake size, just as USB keys bought from a legitimate market are not faked.

 

On SSDs, running this test will needlessly burn write cycles on the SSD. Any defects in the NAND will generally be detected by the drive and mapped out during use. I've seen this in operation and it generally works very well in comparison to hard drives.

 

In addition, SSDs generally use very high grade NAND to begin with, as drives built with faulty NAND (or missing NAND) will panic lock fairly quickly, so there isn't much room to build a low quality drive without causing a riot.

[canthearu]

Just as an additional note, the wear leveling on SSDs ensures all the NAND is used at an equal rate, so the first 128gb of writes on a drive with 128gb of NAND will test the all the NAND in the chips.

 

There is no need to manually run your own test for it.

[Nec_V20]

What you wrote is one of the main uses for H2testw however it is a good tool to use to see if the brand new SSD you bought will write to and read from it without errors.

 

If I buy a brand new SSD and the utility came up with errors then you can be certain I would return it to be replaced.

 

Especially if you are considering RAID 0 you want to be sure that all of the drives in it are error free before you commit your data to them.

 

With regard to, "needlessly burn write cycles on the SSD", let's just work out how much of sacrifice this exacts from the lifetime of the 128 GB you mentioned.

 

I will factor in a ridiculous write amplification factor of ten times (10x) for this calculation and assume that the MLC NAND used has a P/E cycle rating of 5,000 then the total amount of data that could be written to the SSD before it is knackered is (this is of course assuming there is no slack space):

 

(128GB*5,000)/10 = 64 TB

 

So the amount of "damage" inflicted on this by testing the drive completely for writes and reads would be:

 

128GB/64TB*100 = 0.1953125%

 

To put this into perspective you could run this test on the SSD every day for 512 days or 1.4 years before it was consigned to Boot Hill (couldn't resist the pun).

 

That's a price I am willing to pay for the peace of mind that the SSD is functioning as it should from the get go.

[RAM GUY]

I am not familiar with this application but I will ask our engineering about it and let you know. However, we would not be able to use results from this test until it has been accepted by our Engineering. I would suggest using BART HDD Burn in and you will want to secure erase the drive after running the test.

[Nec_V20]

RAM GUY,

 

there is nothing to let me know about. The magazine I mentioned "c't magazin für computertechnik" is so prestigious in Germany that it makes a profit even if they didn't have any advertisements - which makes it unique in the world of print media. It has the nickname "Deutschlands dickstes Männermagazin" (Germany's thickest men's magazine).

 

It was founded in 1983 and I have been reading it since about 1985. This test program has been around for many years and traces its lineage back about two decades.

[canthearu]

What you wrote is one of the main uses for H2testw however it is a good tool to use to see if the brand new SSD you bought will write to and read from it without errors.

 

No it isn't.

 

SSDs do not have a static LBA map, the physical location of each witten sector changes depending on what NAND is currently blank and indirectly by how worn each block is. It would be normal to see a SSD storing the same LBA sector in multiple places, with the older versions waiting for the garbage collector to clean them.

 

In normal use, a SSD controller will use all the least worn blocks first, which means that if you do 128gig of any kind of writes to 120/128gig SSD, it will touch all the NAND cells on SSD at least once.

 

There is no need to explicitly test it as it would be extremely rare for some cells to be bad and the controller not to detect it right away during the write operation.

 

If I buy a brand new SSD and the utility came up with errors then you can be certain I would return it to be replaced.

 

Of course, but this isn't a common failure mode of SSDs, and you arn't testing it any harder than you would by just using it for a few days.

 

The most common mode of failure for SSDs is for the controller to screw up their LBA mapping tables and then fall into an unrecoverable panic lock. This can happen at any time. The second most common mode would be simple electrical failure of the drive, another random event that can happen anytime that pre-testing won't uncover.

 

Especially if you are considering RAID 0 you want to be sure that all of the drives in it are error free before you commit your data to them.

 

If you a considering RAID 0, you will want backups. Not to give yourself a false sense of security by writing a bit of data to the drive and reading it back

 

With regard to, "needlessly burn write cycles on the SSD", let's just work out how much of sacrifice this exacts from the lifetime of the 128 GB you mentioned.

 

Modern Corsair drives use 3000 cycle NAND.

 

Secondly, it isn't so much the size of the sacrifice that is the problem ... as I'd be the first to tell you how durable SSDs are, but the pointlessness of the test that makes it a waste.

 

(I've taken a 128gig MLC NAND drive with 3000 cycles and written close to 1PB data on them before wear induced failure. They last a LONG LONG time)

[Nec_V20]

canthearu,

 

If you a considering RAID 0, you will want backups. Not to give yourself a false sense of security by writing a bit of data to the drive and reading it back

 

Are you seriously trying to tell someone who has worked as Senior German Engineer for Enterprise Disaster Recovery Technical Support and has dealt with over 17,000 individual disaster recovery cases/escalations - only failing to resolve 223 of them (workarounds and kludges don't count) - about the meaning, importance or efficacy of backups? Really?

 

No it isn't.

 

But as I continue to read down your post you contradict yourself and that H2testw actually can uncover an underlying weakness in the SSD one has newly acquired.

 

I very clearly stated that it is a test to see if the SSD is in working order when one newly acquires it NOT that it would be a guarantee with regard to any future functionality.

[RAM GUY]

Okay Guys thanks for the information but I think its best for you to take this off line I am sorry but as I mentioned we cannot officially support any application no matter how it may appear to be with out our engineering's endorsement.

Links was removed.