Technology

Podcast: HDD secure from flash for a decade or extra


On this podcast, we speak to Toshiba’s senior supervisor for enterprise growth in storage, Rainer Kaese, about onerous disk drives (HDDs) and why he thinks their future is assured for a few years to come back.

His arguments centre on the worth differential – with HDD one-seventh of the price of flash – and the way which means flash won’t ever doubtless be used for the largest-scale storage for the foreseeable future.

On the identical time, he says the HDD makers can preserve the hole in worth, for at the least the subsequent decade. 

Antony Adshead: Will flash exchange onerous disk drives?

Rainer Kaese: Most probably no, and by no means.

It’s like tape. It was mentioned that tape would get replaced by different storage applied sciences and tape continues to be round. Onerous disk drives, are and will probably be for fairly a while, the most cost effective approach to retailer on-line knowledge. It’s all about price per capability, and knowledge is rising like hell.

We want financial storage capability, excessive capability or low worth. And in contrast with any flash know-how, onerous disk drives are simply one-seventh the price of every other flash know-how.

Sure, flash could also be far more agile, quicker and better efficiency. However, most often, particularly for actually giant storage [deployments], even onerous disk drives are as quick as we want anyway. And these giant storage [deployments] could be approach too costly; [by] an element of seven in the event that they have been carried out with flash.

So, positively, we are going to merely want onerous disk drives for a lot of extra years to retailer the large quantity of information that mankind is creating.

What’s it about onerous disk drives that ensures its future?

Kaese: Okay, as I mentioned earlier, the foremost factor is price per capability. It’s simply one-seventh [the cost of flash storage].

There’s one other factor. You understand if we have a look at knowledge from the analysts final 12 months, we will see now we have manufactured, offered, put in and stuffed about 1ZB – ie, 1,000EB of HDD capability – however solely about 260EB of SSD capability.

In all of the storage within the enterprises, within the cloud, let’s say 70% to 80% of all knowledge goes to HDDs. Even when flash would catch up in worth at this time, let’s say – and for that, a miracle is required – there’ll for a very long time not be sufficient capability to easily exchange all this 80% HDD base with flash. It wouldn’t work. As of now, [flash is] seven occasions dearer. If this may catch up, it nonetheless wouldn’t work.

And the opposite factor is that arduous disk drives go into the bigger storage [deployments], into the handfuls to a whole bunch of TB as much as a number of PB in a single storage occasion or one storage equipment. Onerous disk drives have been changed by flash [near to] CPUs [central processing units], and we use SSDs [solid-state drives] and NVMe [nonvolatile memory express] in smaller all-flash home equipment the place individuals could make actual good use out of the pace of the SSD.

That is smart, if individuals can use the pace of SSD and monetise the pace of the SSDs. However that is for native storage and for smaller storage [deployments]. For the massive [deployments], it will likely be too costly to make use of all-flash, and it isn’t required as a result of bigger [deployments] indicate now we have many HDDs.

Sure, the HDD is fairly gradual, however many HDDs could be mixed to supply a big efficiency. Not by way of IOPS [input/output operations per second] and agile storage, however you wouldn’t fill 2PB with bank card numbers or invoices. These giant [deployments] are for large knowledge, they’re for machine-generated knowledge, they’re for AI [artificial intelligence] knowledge and for backup.

And, for all these sequential purposes, HDD is quick sufficient to fill any of the widespread community speeds. If the community is 100Gbps, that’s 10GBps, and now we have confirmed in our laboratory {that a} system of 60 onerous disk drives with 1PB to 2PB of storage can fully fill these community speeds.

You possibly can exchange your 1PB to 2PB with all-flash, however then your pockets will probably be fairly empty. You’ll pay 5 occasions to seven occasions the worth, and also you wouldn’t acquire something as a result of it’s restricted by the community. So, for giant storage [deployments] we are going to want onerous disk drives for the foreseeable.

For a way lengthy will onerous disk drives endure towards strong state?

Kaese: The mechanism to maintain the worth distinction to flash-based storage is that now we have to develop onerous disk drives with ever-higher capacities and preserve the price down. The best way it labored up to now and the best way it’ll work for at the least 10 extra years is that we’ll include 2TB or 4TB extra for a similar manufacturing price, and at last additionally the identical worth because the earlier mannequin. That retains the worth erosion in price per capability and this retains the hole to flash.

Costs are eroding, costs are happening for flash, but in addition for onerous disk drives as now we have ever-higher capacities for a similar price versus worth.

So long as we as an business handle to do this, we could have onerous disk drives.

From what I do know is occurring in our laboratories, in our manufacturing and preparation for manufacturing, I can say that as of at this time, we might most likely launch a 40TB onerous disk drive, however it will price a number of thousand {dollars}.

Why would somebody purchase a 40TB onerous disk drive for a number of thousand {dollars} if they’ll purchase two 24TB drives for a number of hundred {dollars}? Launching new know-how for the next worth per TB wouldn’t work.

We now have to launch new applied sciences for a similar worth per TB, and it’s fairly clear that for twenty-four, 28, 30-something – 40, most likely as much as 50 – we will try this. Already, the applied sciences for which have completed their analysis part. They’re now in manufacturing. We now have to convey the manufacturing price down. We now have to determine the economics round it. So, we will say that for the subsequent, say, 10 years, onerous disk drive is secure.

The opposite factor that’s technically doable could be a tough disk drive of 100TB to 200TB. That’s confirmed by analysis. So, 100TB to 200TB could be bodily doable within the present type issue.

What we don’t know but is whether or not we will manufacture it for the worth the market will settle for. If we manufacture it for a similar worth as an SSD, it will likely be too excessive. There must be the hole, and onerous disk drives could slowly part out sooner or later sooner or later, after we can not handle to comply with that development, say, greater capability or the identical price.

However it’s not but recognized if it will ever occur. And in addition, there are guys on the market who say [for] flash know-how, because of its pace, you are able to do far more compression and deduplication; different options.

And in addition, some individuals say flash, though it’s the next capex initially, [by a] issue [of] seven, you’ll save on opex.

Properly, in the mean time, we’re at issue seven. There’s an extended approach to come to an element of 1.5 to issue 2.

And in addition, [there’s] the argument with the saving on opex. We now have finished some calculations. With a capex error of issue seven, the opex financial savings by way of datacentre area are a lot smaller. [There are] financial savings on vitality consumption and cooling – SSD-based storage consumes a little bit bit much less vitality and could be extra effectively cooled – [but this] amortises in 15 to 50 years, relying on which costs you assume. Let’s say 20, 25 years. So, that’s a no brainer at this time.

You wouldn’t save with any kind of flash towards HDD, except you should use its excessive efficiency. This suggests smaller storage, and that is what individuals are doing at this time.

The bigger [deployments] from all arguments will keep for a very long time on spinning disks.