What a Difference Over Provisioning Makes

vbimport

#1

I just posted the article What a Difference Over Provisioning Makes.

Here at Myce.com the Hardware Review Team has recently been considering how best to review and test Enterprise level Solid State Storage solutions. Given our track record in the review and testing of Consumer SSDs we soon found ourselves wondering how Consumer SSDs would perform in tests aimed at Enterprise solutions. We also noticed that Enterprise solutions typically have a much larger amount of NAND storage dedicated for use by their controller (the controller is the heart of an SSD and controls the reading, writing and placement of data in the NAND storage together with the mapping of the storage to a user’s host environment and file system). So we also wondered if consumer SSDs would perform better in enterprise tests if we supplemented the relatively small amount of NAND they have set aside for their controllers by leaving space unassigned for use by the host and its file system. We present our findings, which we feel you may find surprising as well as interesting, in this article.

Click to read the full article here: [http://www.myce.com/review/what-a-difference-over-provisioning-makes-4-66841/](http://www.myce.com/review/what-a-difference-over-provisioning-makes-4-66841/)

Feel free to add your comments below. 

Please note that the reactions from the complete site will be synched below.

#2

What an informative article. I never knew SSDs even offered such an option; I thought OP was hard-set for every drive. Great job. :clap:


#3

Interesting article!


#4

Wow.
Garbage collection algorithms have some serious development potential.


#5

By recording the disk I/O in this manner you are also picking up other traffic, such as TRIM commands and handshaking operations. The disk transfers category includes this ‘unseen’ layer of I/O.


#6

Measuring the disk transfers with this tool is also picking up other disk traffic, such as the TRIM commands and the other underlying I/O from handshaking operations. Unfortunately it invalidates the test results, but the method of charting the data is very nice.


#7

[QUOTE=raidlover;2684602]Measuring the disk transfers with this tool is also picking up other disk traffic, such as the TRIM commands and the other underlying I/O from handshaking operations. Unfortunately it invalidates the test results, but the method of charting the data is very nice.[/QUOTE]

Hi raidlover,

Welcome to the forum.

Would there be much ‘other disk traffic’? The drives were all tested as spares, not as boot drives. There would have been no trim commands sent to the drive during the IOmeter runs.

I agree that it would have picked up other traffic to the logical drive, but the amount of other traffic was, I believe, insignificant.

Regds, JR


#8

The vector in particular is wildly skewed. This is not reproducible. Our internal testing puts that SSD at roughly 44K in steady state when we replicate these conditions.


#9

[QUOTE=raidlover;2685108]The vector in particular is wildly skewed. This is not reproducible. Our internal testing puts that SSD at roughly 44K in steady state when we replicate these conditions.[/QUOTE]

Hi raidlover,

Puzzling

We replicated the result on two separate systems.

I suspect the difference will be caused by an environmental factor. For example, did you run with C1E disabled in your BIOS? Did you run with the Vector plugged into a native Intel Sata 6GBs port?

It would be great if you could let us know the spec of your test system.

Regds, JR


#10

Hi raidlover,

For interest I have repeated the 4K Write Saturation Test on a Vector 256GB using our Oakgate Test Unit. Here are the results, firstly - allowing writes to 60% of the LBA (i.e. with a large additional OP) and secondly - allowing writes to 100% of the LBA (i.e. no additional OP)

Interestingly this shows that a steady state has not quite yet been reached after 45 minutes. It also shows that the IOPS norm has not fallen below 60k - but still significantly faster than the 44K you are observing. I suspect the differences between what we see on the Oakgate test Unit and the results observed in our article’s Windows based tests can be explained by the Intel RST driver and its caching capabilities.

You’ll also notice that a regular, temporary dip in IOPS can be clearly seen. I suspect that this is where the controller is quickly moving data from ‘Performance NAND’ to ‘Storage NAND’ (which is a feature of recent Indilinx firmware) to free up Performance NAND to accommodate further writes.

You can see here that even with no additional OP steady state has not been achieved after 45 minutes when tested in the Oakgate Unit.

I should have tested for longer :slight_smile:

An observation raidlover - I warmly welcome people challenging our test results (it is very healthy) but perhaps a challenge that is used with absolute assertions such as ‘invalidates, widely skewed and/or not reproducible’ should first be considered more carefully.

Kind regards, Jeremy


#11

…and for good measure here is a two hour run on the Oakgate Unit, showing that, with a large additional OP, the Vector settles down at a norm above 60k -

Regds, JR


#12

I noted that you are testing with a SATA port. There are very few SATA (2-3) ports on server motherboards. Does your Oakgate unit have a different means of connection? HBA, RAID controller, etc, as would be used in an actual enterprise environment.
I also note that the results are 10K lower than the methodology used with a consumer motherboard, which is a rather large difference.


#13

[QUOTE=raidlover;2685237]I noted that you are testing with a SATA port. There are very few SATA (2-3) ports on server motherboards. Does your Oakgate unit have a different means of connection? HBA, RAID controller, etc, as would be used in an actual enterprise environment. .[/QUOTE]

Of course. Are you now questioning the efficacy of an Oakgate Test Unit? Come on m8 you can do better than that :slight_smile:

[QUOTE=raidlover;2685237]I also note that the results are 10K lower than the methodology used with a consumer motherboard, which is a rather large difference.[/QUOTE]

Clearly.

What do you think might explain the difference?

Please let us know what environment you used to get 44K. Maybe there is a clue here :slight_smile:

Regds, JR


#14

[QUOTE=JReynolds;2685245] Are you now questioning the efficacy of an Oakgate Test Unit? Come on m8 you can do better than that :slight_smile: [/QUOTE]

Not questioning the efficacy of the unit, I am just not sure of its capabilities.
No need to “do better than that”. Is there a unit model # so i can give it a look, sounds interesting.

[QUOTE=JReynolds;2685245]
What do you think might explain the difference? :slight_smile:

Please let us know what environment you used to get 44K. Maybe there is a clue here :slight_smile:

Regds, JR[/QUOTE]
Testing was with an HBA. Any 6Gb/s HBA will do. I would be interested to see the results on an HBA or RAID controller, it should vary from the SATA results significantly. :slight_smile:


#15

[QUOTE=raidlover;2685247]Not questioning the efficacy of the unit, I am just not sure of its capabilities. [/QUOTE]

Very wise :slight_smile: We have a non-standard unit. Please see www.oakgatetech.com. I’ll soon be publishing an article to document our enterprise testing methodology using the Oakgate unit. By the way, it is the same unit that is used by many of the world’s leading enterprise SSS solutions manufacturers.

[QUOTE=raidlover;2685247]Testing was with an HBA. Any 6Gb/s HBA will do. I would be interested to see the results on an HBA or RAID controller, it should vary from the SATA results significantly. :)[/QUOTE]

Understood. By the way the results published in the article for the OCZ Vertex 4 were run with the V4 plugged into an HBA (LSI 9207 8i in my desktop as detailed below) - so I guess you get similar results if you have a V4 to hand?

Oh - and there was a big clue then.

Regds, JR


#16

I wish I had a Vertex 4 :frowning:


#17

Thanks for this, I learned a lot. I’m still debating Partitioning Usefulessness, and the whole To Scan Or Not To Scan issue. Now I’ve got OP to play with, too.


#18

To Scan or not to scan?


#19

Optical disks.


#20

I just ordered a new Seagate 600 Pro 400gb SSD, it comes from the factory 22% Over provisioned. I believe its really a 512 gb SSD. But after formatting only 376gbs. Its a Mid range Enterprise SSD good 1Petabyte or 5 years of writes, probably last me forever. LOL:bigsmile::wink: