News Firm Claims HDDs Can Use Less Power Than SSDs

[Admin]

Storage provider Scality compares the power efficiency between high-density HDDs and QLC SSDs.

Firm Claims HDDs Can Use Less Power Than SSDs : Read more

 

[vehekos]

This is highly dubious. Which HDD models were compared to which SSD?

What do they mean by power density? energy per Mb, or "power" by Mb. If you consume the same energy in less time, you had higher power, but the same energy.

 

[TMTOWTSAC]

This article is incredibly disappointing. Scality didn't bother to test or benchmark anything. They say so right in their article.

For each workload profile, drives are assumed to be in the specified power state for the percentage of time indicated.

They took power draw numbers for idle/read/write from the manufacturers and multiplied them by the time percentages. That's it. They didn't make any power measurements. Only 2 drives were compared. And worst of all, they normalized against time, not workload.

They imagined a theoretical scenario in which both drives spent the same amount of time reading/writing/idling. In the real world, drives go back to idling once they finish an operation. The faster drive will spend far more time at idle. And that isn't going to be the mechanical drive.

 

[Alvar "Miles" Udell]

So it's 5 day old "news" at this point, already been copypasted on another website, and is so dubious that it shouldn't even be posted at all, so why is it being posted about on TH?

 

[Johnpombrio]

Who cares? HDDs are SLOW and noisy. Unless I purposefully keep them spinning, they spin down and make a chatter when they are woken up. I have had several HDDs fail and only one early SSD. I get that they are trying to keep their business alive, but the only reason for HDDs is for large low access storage for less cost, and that is shrinking every year.

 

[Vanderlindemedia]

Who cares? HDDs are SLOW and noisy. Unless I purposefully keep them spinning, they spin down and make a chatter when they are woken up. I have had several HDDs fail and only one early SSD. I get that they are trying to keep their business alive, but the only reason for HDDs is for large low access storage for less cost, and that is shrinking every year.

If you run a cluster with i dont know, 128, 256, 512 HDD's large, 5W per disk is a big difference. Not everything is based around speed or required absolute speed.

 

[Carson365]

Clearly I'm not the only one who noticed the gaping holes in this comparison. Whether or not the data was gathered from specs rather than experimental observation as another commenter stated, it's a flawed comparison in the first place. Why measure the power consumption of drives meant for different use cases under the same useless condition? If an SSD draws twice the energy but it does so in a quarter of the time, it's used half of the energy that the HDD would to do the same thing.

 

[Carson365]

If you run a cluster with i dont know, 128, 256, 512 HDD's large, 5W per disk is a big difference. Not everything is based around speed or required absolute speed.

But the speed also affects the power draw. It's why utility companies measure in Kilowatt Hours, KwH. If your 5W per disk are being drawn for 20s instead of 1m, that's still an overall savings.

 

[bit_user]

This is highly dubious. Which HDD models were compared to which SSD?

It's right in the article:

"Scality used the Micron 6500 ION 30.72TB QLC SSD and the Seagate Exos X22 22TB 7,200 RPM hard drive"
​

These datacenter SSDs aren't like the ones in your PC. They have disproportionately higher idle power, which really swings the comparison with HDDs.

What do they mean by power density? energy per Mb,

This. In datacenters, people talk about density all the time. So, I'm sure they mean power normalized by capacity.

The source is the most suspicious part. Scality.com describes them as an "Object storage & Cloud Solutions Leader". I think that "publication" which put out the study is basically a PR arm that they use to drive positive press coverage for themselves and their key customers.

 

[bit_user]

Why measure the power consumption of drives meant for different use cases under the same useless condition?

Probably because HDD companies are starting to get a bit concerned about QLC SSDs encroaching into more HDD use cases and market niches.

On the horizon, we can foresee a point where SSDs can be competitive purely on the basis of GB/$, even in cases where their performance isn't required. In such a situation, normalizing by performance would be misguided.

If an SSD draws twice the energy but it does so in a quarter of the time, it's used half of the energy that the HDD would to do the same thing.

True. They should've adjusted the duty cycle estimates to account for the performance differences.

 

[derekullo]

Simply calculating how many watts it takes to transfer 500 gigabytes of data would settle the issue.

Something like

The Best NVMe SSD for Laptops and Notebooks: SK hynix Gold P31 1TB SSD Reviewed

www.anandtech.com

 

[InvalidError]

Simply calculating how many watts it takes to transfer 500 gigabytes of data would settle the issue.

For an online storage provider, absolute speed of a single transfer stream or drive isn't much of a factor as most clients won't be accessing online storage at connection speeds that will pose any issue for HDDs, especially after software-RAID scattering data cross multiple drives for resiliency. They are more concerned about the overall cost of keeping a given amount of storage online.

 

[USAFRet]

For an online storage provider, absolute speed of a single transfer stream or drive isn't much of a factor as most clients won't be accessing online storage at connection speeds that will pose any issue for HDDs, especially after software-RAID scattering data cross multiple drives for resiliency. They are more concerned about the overall cost of keeping a given amount of storage online.

Exactly.

But for thee and me in our living rooms, completely different.

 

[Co BIY]

Do hard drives go into a sleep mode when not accessed in data centers ?

I can imagine that a great deal of archived data could be stored on "non-spinning" Iron very efficiently. Although probably SSD's in a "suspended" state should be able to do the same. A lot of stored data is rarely accessed.

 

[InvalidError]

Do hard drives go into a sleep mode when not accessed in data centers ?

Depends on what they are being used for. In the case of an online storage company, having thousands of customers reading and writing data striped across multiple drives means drives rarely are idle long enough to bother spinning down.

Add data scrubbing during otherwise idle time to the mix to catch bit rot before it may cause data loss, I doubt HDDs get much down-time.

 

[bit_user]

Do hard drives go into a sleep mode when not accessed in data centers ?

In some "near-line" storage applications, I would expect so. Backups might be one example, but also data which is not expected to be accessed very frequently. I'm not speaking from any first-hand knowledge, however.

I can imagine that a great deal of archived data could be stored on "non-spinning" Iron very efficiently. Although probably SSD's in a "suspended" state should be able to do the same. A lot of stored data is rarely accessed.

Yes, good point about non-spinning HDDs.

As for the SSD equivalent, I don't know exactly what this "suspended" state is. The datasheets I've seen of datacenter SSDs (and I've looked at a few of them; just bought one) don't specify anything like that. I gather you mean they're effectively powered-down, but the only way I know to do that would be to put the host machine in a suspended state. That's trickier proposition, as it will be hosting lots of drives and I'd imagine it wouldn't go too long before something on one of them needs to be accessed.

 

[SyCoREAPER]

It's doesn't mention what they consider a read. Is it the drive just reading the allocation?

Speaking to larger files, It will take an HDD probably three times longer to read the same sized file resulting in 3 times more power needed.

 

[bit_user]

Depends on what they are being used for. In the case of an online storage company, having thousands of customers reading and writing data striped across multiple drives means drives rarely are idle long enough to bother spinning down.

I could imagine data either being sorted a priori or dynamically into a near-line storage tier. Something like backups or video surveillance might be good examples of cases where you know data is unlikely to be read. So, once you fill a drive with such data, you can explicitly spin it down, or just rely on an automatic spin-down based on the amount of time since last access.

As for striping, various things I've read about cloud storage suggests they prefer replication over RAIDs. It probably depends on the application, but rebuild times for RAIDs are getting so long that the chance of another failure occurring during the process becomes non-trivial. Also, RAIDs are good for sequential access patterns & throughput, but rather poor for random ones. The larger the array, the worse it gets. So, if you use large arrays to amortize the overhead of RAID-6, then it'll really hit your IOPS, compared to what you can get with JBOD + replication.

Add data scrubbing during otherwise idle time to the mix to catch bit rot before it may cause data loss, I doubt HDDs get much down-time.

The typical heuristic for scrubbing is usually about once per month. However, it'll take a few days, if you do it at full speed. Dual-actuator drives would cut that in half, but they're still far from the norm.

 

[TerryLaze]

With hardware, companies have to plan for equipment utilization rate, if you use your hardware for less than x% of the time you are losing money but you also have to leave some amount of headroom for rush-hours or emergencies.
They have to calculate based on how many hours on average their storage (or anything) will be running and not based on a single workload that might be run once in a full moon.
They expect the storage to run non stop for so many "percent of hours"

What are the benefits and challenges of using the equipment utilization rate formula?

Learn what the equipment utilization rate formula is, how to apply it in different contexts, and what are its benefits and challenges.

www.linkedin.com

 

[Ewitte]

It's doesn't mention what they consider a read. Is it the drive just reading the allocation?

Speaking to larger files, It will take an HDD probably three times longer to read the same sized file resulting in 3 times more power needed.

Or 100 times if it's highly random.

 

[Viking2121]

Sure, the only time I can see an SSD using more power is during writes, but they are so fast, the HDD will sit there and consume power for much longer as the SSD will sit there idle doing nothing because the task is already done.

 

[InvalidError]

Sure, the only time I can see an SSD using more power is during writes, but they are so fast, the HDD will sit there and consume power for much longer as the SSD will sit there idle doing nothing because the task is already done.

If you are building an online storage service, most of the storage will be idle most of the time even if it is only lowly HDDs.

 

[Viking2121]

If you are building an online storage service, most of the storage will be idle most of the time even if it is only lowly HDDs.

The power draw the drive needs to spin up still far exceeds what an SSD will use waking up from idle or low power state, No matter how anyone looks at it, an SSD will always consume less, unless you look at the numbers at the right time and at the right work loads as such int he that link.

 

[PEnns]

Yeah, sure.

Even if it's true, Guess what? People are not going to dump their SSDs and go back to spinning disks.

 

[bit_user]

Sure, the only time I can see an SSD using more power is during writes, but they are so fast,

Not sure if some people skipped the article and went straight to the comments, but they took power usage directly from the drive manufacturer (Micron):

Power Data Per Drive	SSD	HDD	HDD Advantage
Idle (watts)	5	5.7	-14%
Active read (watts)	15	9.4	37%
Active write (watts)	20	6.4	68%
Read-intensive workload (avg. watts)	14.5	8.7	40%
Write-intensive workload (avg. watts)	18	6.6	63%
Power-density read-intensive (TB/watt)	2.1	2.5	19%
Power-density write-intensive (TB/watt)	1.7	3.3	94%

Even at idle, the SSD's power usage is non-trivial. I've looked at datasheets, from several datacenter SSDs, and this is the lowest idle power I've seen yet. These are not the same as SSDs that you find in even a 1337 G4M3R PC, so set aside what you think you know about SSD power consumption.