SSDs Kill The 15K HDD, Seagate Rolls Out Last Generation

Status
Not open for further replies.

takeshi7

Honorable
Nov 15, 2013
105
3
10,685
One of the advantages 15K HDDs have over SSDs that you failed to mention in the article is the write endurance is much higher for an HDD.
 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
858
315
19,360


Well, here is an oddity... that common belief is both right and wrong.

It all boils down to how much data you can actually write to the drive. This is called usable endurance. HDDs are so slow that they cannot write enough data over their lifespan to compete. Here is an excerpt from another site where I wrote a post covering this topic in 2013. The numbers are probably even more in favor for SSDs with the newer models.

"While crunching the numbers it occurred to me that endurance isn't quite as clear-cut as one would expect. Taking a look at an entry-level SSD we see that an endurance of 3 DWPD (Drive Writes Per Day) is pretty standard for a random workload. The sequential workload endurance can be much higher, from 7-10 DWPD, depending upon the SSD. While the glass platter substrate of an HDD has a very high tolerance for heavy workloads, the mechanics of the moving parts conspire to hinder the speed, and thus the useable endurance of the drive. In our testing, the fastest 15K HDDs write at a speed of roughly 450 IOPS for both 4K and 8K random write workloads. For 4k this equates to roughly 148 GB's of data written per day. For 8K access, common to many server workloads, we arrive at roughly 296 GB of potential data written per day.

An entry-level SAS SSD, by comparison, can provide over 30,000 IOPS of 4k write speed and 18,000 8K write IOPS in steady state. This equates to nearly 9.6 TB of potential data writes per day for 4K access, and 11.5 TB's per day for 8K write access. This is a huge advance over the HDD, and could necessitate throttling of the SSD to keep it within the expected warrantied workload of 3 DWPD. With the capacity of new SSDs touching 2TB, this can provide up to 6 TB of useable endurance per day for the SSD, in comparison to the lowly 148-296 GB attainable by today's fastest HDDs.

The 2TB SSD can provide 10,950 TBs of 8K write activity (endurance) over the warrantied period of five years, compared to the HDD with 534 TBs.

The slower speed of the HDD negates its ability to take advantage of its nearly-unlimited endurance, while the SSD can write nearly 20X more data over five years.
"


In a nutshell, SSDs offer up to 20X (and even more) usable endurance than HDDs.





 

PC-Cobbler

Distinguished
Jun 12, 2015
47
0
18,530
Seagate buy Micron? Not likely, given that Micron is bigger in every monetary respect. SK Hynix would be a possibility, but South Korea's chaibol culture might prevent that. I think it's much more likely for Seagate to partner with a Chinese fab, given its close ties to China (Seagate's job listings often ask for people who speak fluent Mandarin). Another possibility is Seagate being bought by a Chinese company such as Tsinghua or XMC.
 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
858
315
19,360


Tsinghua has already tried, for the record, to get their hooks into Micron. They are the persistent sort, so may try again after the failed bid to get SanDisk by buying into WD. http://www.tomsitpro.com/articles/wd-unisplendour-investment-acquisition-sandisk,1-3142.html

This is also interesting - http://www.tomsitpro.com/articles/wd-unisplendour-jv-nand-fab,1-3015.html

Micron has already taken measures to avoid a buyout from Tshinghua/XMC (now operating under the name Wuhan Xinxin Semiconductor Manufacturing). Read more here - http://www.tomshardware.com/news/samsung-ssd-flash-micron-china,32347.html

Yes, the real surprise would be if Micron bought Seagate, which might be a possibility as it still hasn't finished the Inotera acquisition, which would likely leave it in too much debt to do so at favorable terms.

I'm thinking merger. The present strategic alliance already has the two companies operating pretty closely together.

i don't think XMC (et.al,) is interested in Seagate. Seagate with a NAND fab; yes. Seagate as-is; no way (imo). Tsinghua/Unisplendour is angling for NAND fabs, not spinning rust.

(also, you did a good job of encapsulating the SK hynix issue-my thoughts exactly. SK hynix is too big, and I doubt they will sell off their strategic NAND assets)
 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
858
315
19,360

Gah. Fixed.
 

takeshi7

Honorable
Nov 15, 2013
105
3
10,685


OK, that makes sense. I will revise my original statement to HDDs have vastly higher sequential write endurance.

 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
858
315
19,360


Actually, that might not be true, though I would have to work the math out...but enterprise SSDs are much much faster with sequential data, and SSD endurance increases by a factor of 7 to 10x with sequential workloads. So, nope. HDDs still lose.

 

takeshi7

Honorable
Nov 15, 2013
105
3
10,685


Tom's Hardware should do a test of this. Sequential write the entire LBA space on an HDD and an equally sized SSD until they both die. With non compressible data. I would love to see the results.
 

mordac35

Distinguished
Oct 8, 2009
5
0
18,510
I am waiting for the day when SSD's can replace my HDD's in my NVR. 100TB is still prohibitively expensive in an SSD.
 

TadashiTG

Honorable
Apr 23, 2013
80
0
10,640


HDD too slow to complete the test in a reasonable time(decades to finish?).
 

takeshi7

Honorable
Nov 15, 2013
105
3
10,685

As long as it's sequential the HDD shouldn't be too slow compared to the SSD. Maybe 2-3x slower than a SATA SSD. I still believe in the end the HDD would write more data than the SSD.


 

bit_user

Polypheme
Ambassador
Thanks for the article, Paul. I'm sure future historians will appreciate it.

I'm just glad to see a HDD finally exceed SATA/SAS 2 speeds.

Ah, but if you actually tried to write 8k random writes, nonstop, then I'm pretty certain the drive would die well before the end of the warranty period.

However, if you went back and added some larger writes, the ratio would quickly approach 1:1.

Nice try.
 

bit_user

Polypheme
Ambassador
But you also aren't running 100 TB of 15 kRPM disks. Those aren't that much cheaper per-GB than SSDs (which is why they're the first casualty).
 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
858
315
19,360


Sorry, I probably wasn't clear enough, as that was a copy/paste from an old article (which is much longer). However, the relevant bits are there and apply to warrantied writes only.

First, we used verified performance test data from our lab (in a steady-state) to show the maximum speed;

This equates to nearly 9.6 TB of potential data writes per day for 4K access, and 11.5 TB's per day for 8K write access

Second, I accounted for the fact that most enterprise SSDs throttle to keep you within the warranty period. So, I narrowed the numbers down to only the warrantied endurance level (which is why these metrics are lower than the 9.6TB and 11.5TB listed above);

This is a huge advance over the HDD, and could necessitate throttling of the SSD to keep it within the expected warrantied workload of 3 DWPD. With the capacity of new SSDs touching 2TB, this can provide up to 6 TB of useable endurance per day for the SSD, in comparison to the lowly 148-296 GB attainable by today's fastest HDDs.

Then we used the warrantied endurance only to compile the final calculations. All of this data would be written within the warranty.
 

takeshi7

Honorable
Nov 15, 2013
105
3
10,685
This is a huge advance over the HDD, and could necessitate throttling of the SSD to keep it within the expected warrantied workload of 3 DWPD. With the capacity of new SSDs touching 2TB, this can provide up to 6 TB of useable endurance per day for the SSD, in comparison to the lowly 148-296 GB attainable by today's fastest HDDs.
I don't know where you got that figure saying the fastest HDDs can only do 148-296 GB per day. The Seagate SkyHawk is rated for around 500 GB/day (180 TB/year), and I'm sure in reality the amount of writes the drive can handle is much higher since it doesn't have to wear level any flash. Paul, you should actually test this instead of just talking a bunch of nonsense about it.

source: http://www.seagate.com/www-content/product-content/skyhawk/files/skyhawk-ds-1902-3-1608us.pdf


 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
858
315
19,360


For sequential data, yes, HDDs can write more data per day. However, we are discussing a competition between SSDs and HDDs - and data centers buy SSDs for perforamance-oriented workloads, ie, random data. HDDs cannot write that much random data per day by any measure, even if you multiply its performance by a factor of ten.

Here is the state of the 7,200RPM-class you are referring to, but these are enterprise HDDs, thus faster. http://www.tomsitpro.com/articles/seagate-8tb-enterprise-capacity-3.5-hdd-v5-review,2-17-3.html

Testing? Here is the current state of HDD performance with the leading 15K HDDs (which are the fastest). This includes every relevant drive on the market. I've tested them for years, so you can google and find many many more articles over the years. This is the most current state of the market. (though I do have ten of the new 15k.6 Seagate drives in the lab, and will provide an update to this test pool shortly.)

http://www.tomsitpro.com/articles/hgst-ultrastar-c15k600-hdd,2-906-4.html


For SSDs, here is the current state of the SATA SSD market, which is also a segment I have tested for years. Also, bear in mind these are SATA SSDs that are on the low end of both endurance and performance, and they still beat HDDs handily in the amount of usable performance they provide.

http://www.tomsitpro.com/articles/intel-dc-s3610-800gb-enterprise-ssd,2-924-3.html

As to sequential data, if you feel the need to work out the math, you will find a very similar situation.

So, yes, this is based upon actual testing. The rest is just math. Perhaps I should note that SSD vendors have used this same line of reasoning for years when selling SSDs, but those years began after my 2013 article :)


 

takeshi7

Honorable
Nov 15, 2013
105
3
10,685


So how much sequential data did the 15K HDD write before it died, and how much sequential data did the SATA SSD write before it died? And what about the sequential write endurance for equal capacity HDDs and SSDs, and the sequential write endurance for equally priced HDDs and SSDs? Your testing doesn't seem to answer these important sequential write questions.
 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
858
315
19,360


HDDs, and SSDs, can live long beyond the warranty period, but that is irrelevant. Data centers rarely use tech longer than five years -- it isn't like they've got PATA drives still spinning.

We are discussing, specifically, the five-year warrantied period, as it is the only measurement that matters.

As noted, an SSDs sequential write endurance is much higher than random write endurance, which is a common bit of knowledge. We show the worst-case SSD performance in sequnetial workloads in those links above, and also provide the best-case HDD speed for sequential data as well (all you have to do is go to the pages labeled (128KB Sequential yada yada).

HDDs have workload restrictions, simply compare them to the warrantied sequential write endurance of an SSD, and there you have your answer. If we were to run an endurance test of sequential data on an SSD it would easily take more than five years. They are warrantied for five years of RANDOM data, which is exponentially rougher on the NAND (kills endurance faster) than sequential data. A sequential data test would take 10 or more years, likely. None of that matters, however, as it is only warrantied for five years. I have provided links to more than enough data to do some simple math and extrapolate the metrics. I am interested in your findings, feel free to post.

 

takeshi7

Honorable
Nov 15, 2013
105
3
10,685
I have provided links to more than enough data to do some simple math and extrapolate the metrics. I am interested in your findings, feel free to post.
I am honored that a technology review site has asked one of their readers to perform this important test instead of performing it themselves. Unfortunately I don't have the resources and access to components that Tom's Hardware has, so can I borrow some drives for this? Otherwise I will need to do this on a budget, and I will compare the sequential write endurance of two similarly cheap drives. Probably a 1TB HDD and a 120GB SSD. Paul, do you know of any programs that can sequential write non-repeating, non-compressible data across the entire LBA range in a loop, and keep track of the amount of data written?

 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
858
315
19,360


Sigh. I am not asking you to test anything, I am suggesting that you use existing performance data and figure out the difference via basic math, as I don't have a spare ten years to see how long it would take to kill an SSD with sequential data.

Iometer, fio, iozone, etc, can accomplish the task, if you choose to accept it.
 

takeshi7

Honorable
Nov 15, 2013
105
3
10,685


Extrapolating with math isn't the same thing as actually testing it. If it were Christopher Columbus would have actually made it to India. I don't know of any journalists that have actually tested the write endurance of HDDs, and the only real test I know of for real SSD write endurance was the one by The Tech Report, but that wasn't done at the LBA level.
 
Status
Not open for further replies.