SSD Extreme Overprovisioning = 5X Intel IOPS/Endurance???

Silencing hard drives, optical drives and other storage devices

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

SSD Extreme Overprovisioning = 5X Intel IOPS/Endurance???

Post by ces » Sat Apr 30, 2011 8:58 am

ilovejedd wrote:
Eunos wrote:The 120GB G2 is a value sweet-spot in the meantime.
+1. Couldn't help but wish I bought more of them during Black Friday when they were going for $200 ($165 after MIR) a piece. I only bought two (the limit for the rebate) as I was expecting the G3 drives to arrive February. Now, we're already into March and still no signs of cheaper 25nm G3 SSD's and I need 4 more of these things. :roll:
I ended up deciding to get two 120G G2s. They should meet my needs for some time. I am just going to use them as boot drives.

I am going to use 50% of the 120G to over provision them. This should increase iops by over 5x and its lifetime by over 5x.... at least according to Intel. That doesn't seem right, but that is what they say here. See:
http://cache-www.intel.com/cd/00/00/45/ ... 459555.pdf

A five times increase in IOPS seems overly much. Am I not reading that right. It seems too easy. Does anyone know what I should expect from such a jump in just the Iops attribute of performance?

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

SSD Overprovisioning

Post by ces » Tue May 03, 2011 8:33 am

Is there anyone with experience with SSD over provisioning?

dhanson865
Posts: 2198
Joined: Thu Feb 10, 2005 11:20 am
Location: TN, USA

Re: SSD Overprovisioning

Post by dhanson865 » Tue May 03, 2011 9:21 am

ces wrote:Is there anyone with experience with SSD over provisioning?
Yes

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Re: SSD Overprovisioning

Post by ces » Tue May 03, 2011 1:29 pm

dhanson865 wrote:
ces wrote:Is there anyone with experience with SSD over provisioning?
Yes
What did you learn? Was the impact anything like what the Intel white papers says?

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

SSD Extreme Overprovisioning

Post by ces » Thu May 05, 2011 10:45 pm

I guess dhanson865 isn't talking. Maybe he doesn't like my politics :) (see viewtopic.php?f=18&t=61480) Maybe he doesn't have the answer. Maybe for other reasons.

But someone out there... have you done over provisioning? What has been your experience with it?

Eunos
Friend of SPCR
Posts: 378
Joined: Mon Dec 12, 2005 3:29 am
Location: Melbourne, Australia

Re: SSD Extreme Overprovisioning = 5X Intel IOPS/Endurance??

Post by Eunos » Fri May 06, 2011 8:19 am

I don't have overprovisioning experience but can make a couple of comments.

Regarding endurance, it is important to put this in a real-world context by calculating how many writes it will actually do. As far as I know, flash drives all drop dead around the 10-year age, to say nothing of being obsolete before that time. I did zero overprovisioning on my X25-M and the most worn drive still has 98% remaining.

Regarding IOPS, this is the first time I've heard claims of multiplying performance. Even if it's true, again it needs to be put in the context of the actual workload that the drive will be put through. It may not be a noticable gain.

Of course there is nothing wrong with overprovisioning and I would certainly recommend formatting the drive to 5-10% less than its maximum capacity, but that PDF seems to overstate the actual benefits.

andymcca
Posts: 404
Joined: Mon Jan 11, 2010 8:19 am
Location: Boston, MA, USA

Re: SSD Extreme Overprovisioning = 5X Intel IOPS/Endurance??

Post by andymcca » Fri May 06, 2011 9:40 am

Eunos wrote:As far as I know, flash drives all drop dead around the 10-year age
Do have any suggested reading to this end? I'm just curious if it applies to current processes, or if it is historical based on 10 year old processes?

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Re: SSD Extreme Overprovisioning

Post by ces » Fri May 06, 2011 6:21 pm

Eunos wrote:flash drives all drop dead around the 10-year age, to say nothing of being obsolete before that time. I did zero overprovisioning on my X25-M and the most worn drive still has 98% remaining.
I have an OCZ that, if it continue on its current trajectory, looks like it will drop at around 18 or 20 months. It seems like what you are saying may be a fair statement about Intel X25-M drives.... but it appears that it doesn't apply across the board... to my great personal surprise.... because what you say is what I used to think was so. And it likely is so... but not in all contexts or situations.

ces
Posts: 3395
Joined: Thu Feb 04, 2010 6:06 pm
Location: US

Re: SSD Extreme Overprovisioning

Post by ces » Fri May 06, 2011 6:34 pm

Eunos wrote:Regarding IOPS, this is the first time I've heard claims of multiplying performance. Even if it's true, again it needs to be put in the context of the actual workload that the drive will be put through. It may not be a noticable gain.
I would normally discount a claim like that... except it is made by Intel in an official document. And I believe I am reading it right.

It may just apply to Intel SSDS.... more or less. Or it might have more general applicability. It requires too much skill and effort to do a good test. We will have to leave that to Anand and/or SPCR.

But as you indicate... the real question isn't whether or not it can increase IOPS... The real question is if it does.... so what... what is it's real life impact?

With the caveat that I don't feel that I have a good intuitive understanding of cause and effect in start up speed.... It seems to me that a lot of the startup time near the end of the booting process involves the drives scooting around loading a whole bunch of small files. Seems to me that increased IOPS would decrease that bottleneck. I think it probably helps with any poorly written application that is scurrying around attempting to load a whole bunch of small files.

For a desk top user, maybe that is the only benefit. I wouldn't expect that to help if you are loading or moving large files.

You may be talking me out of doing a heavy duty over provisioning on my new Intel SSDs. I am going to have to think about it.

Eunos
Friend of SPCR
Posts: 378
Joined: Mon Dec 12, 2005 3:29 am
Location: Melbourne, Australia

Re: SSD Extreme Overprovisioning = 5X Intel IOPS/Endurance??

Post by Eunos » Fri May 06, 2011 8:19 pm

andymcca: That is one of those 'consensus' things which in practise can be expected to vary either way quite a bit. But here's a couple of references of sorts. Keep in mind SSDs use the same basic technology as USB flash drives - they can withstand more writes but I don't think they would last any longer in shelf storage time.

http://en.wikipedia.org/wiki/USB_flash_drive
"Some allow 1 million write or erase cycles[citation needed] and offer a 10-year[citation needed]shelf storage time."

http://www.ehow.com/about_5393136_life- ... drive.html
"Thus, a USB flash drive keeps data indefinitely, at least for 10 years, without any loss of information."

ces: There was a very interesting review on StorageReview a while back of a mega-expensive PCIe enterprise-grade SSD offering massive IOPS for a ridiculous price. In terms of real-world PC usage, they found it to be not significantly faster than a consumer-grade Vertex 2 or similar. In high queue depths it absolutely blew the consumer drives out of the water.

Example: booting up Windows might be roughly the same for both, but if you had 20 computers' operating systems all on the one SSD and booted them simultaneously, the mega-IOPS drive would be much faster. This is why I say that increasing the IOPS on an X25-M may not be significant in real-world conditions.

Cheers.

Post Reply