In this podcast, we talk to Toshiba’s senior manager for business development in storage, Rainer Kaese, about hard disk drives (HDDs) and why he thinks their future is assured for many years to come.
His arguments centre on the price differential – with HDD one-seventh of the cost of flash – and how that means flash will never likely be used for the largest-scale storage for the foreseeable future.
At the same time, he says the HDD makers can maintain the gap in price, for at least the next decade.
Antony Adshead: Will flash replace hard disk drives?
Rainer Kaese: Most likely no, and never.
It’s like tape. It was said that tape would be replaced by other storage technologies and tape is still around. Hard disk drives, are and will be for quite some time, the cheapest way to store online data. It’s all about cost per capacity, and data is growing like hell.
We need economic storage capacity, high capacity or low price. And compared with any flash technology, hard disk drives are just one-seventh the cost of any other flash technology.
Yes, flash may be way more agile, faster and higher performance. But, in most cases, especially for really large storage [deployments], even hard disk drives are as fast as we need anyway. And these large storage [deployments] would be way too expensive; [by] a factor of seven if they were implemented with flash.
So, definitely, we will simply need hard disk drives for many more years to store the enormous amount of data that mankind is creating.
What is it about hard disk drives that ensures its future?
Kaese: Okay, as I said earlier, the major thing is cost per capacity. It’s just one-seventh [the cost of flash storage].
There’s another thing. You know if we look at data from the analysts last year, we can see we have manufactured, sold, installed and filled about 1ZB – ie, 1,000EB of HDD capacity – but only about 260EB of SSD capacity.
In all the storage in the enterprises, in the cloud, let’s say 70% to 80% of all data goes to HDDs. Even if flash would catch up in price today, let’s say – and for that, a miracle is required – there will for a long time not be enough capacity to simply replace all this 80% HDD base with flash. It wouldn’t work. As of now, [flash is] seven times more expensive. If this would catch up, it still wouldn’t work.
And the other thing is that hard disk drives go into the larger storage [deployments], into the dozens to hundreds of TB up to several PB in one storage instance or one storage appliance. Hard disk drives have been replaced by flash [near to] CPUs [central processing units], and we use SSDs [solid-state drives] and NVMe [nonvolatile memory express] in smaller all-flash appliances where people can make real good use out of the speed of the SSD.
That makes sense, if people can use the speed of SSD and monetise the speed of the SSDs. But this is for local storage and for smaller storage [deployments]. For the large [deployments], it will be too expensive to use all-flash, and it is not required because larger [deployments] imply we have many HDDs.
Yes, the HDD is pretty slow, but many HDDs can be combined to provide a significant performance. Not in terms of IOPS [input/output operations per second] and agile storage, but you wouldn’t fill 2PB with credit card numbers or invoices. These large [deployments] are for big data, they are for machine-generated data, they are for AI [artificial intelligence] data and for backup.
And, for these types of sequential applications, HDD is fast enough to fill any of the common network speeds. If the network is 100Gbps, that’s 10GBps, and we have proven in our laboratory that a system of 60 hard disk drives with 1PB to 2PB of storage can completely fill these network speeds.
You can replace your 1PB to 2PB with all-flash, but then your wallet will be pretty empty. You will pay five times to seven times the price, and you wouldn’t gain anything because it’s limited by the network. So, for large storage [deployments] we will need hard disk drives for the foreseeable.
For how long will hard disk drives endure against solid state?
Kaese: The mechanism to keep the price difference to flash-based storage is that we have to develop hard disk drives with ever-higher capacities and keep the cost down. The way it worked in the past and the way it will work for at least 10 more years is that we will come with 2TB or 4TB more for the same production cost, and finally also the same price as the previous model. That keeps the price erosion in cost per capacity and this keeps the gap to flash.
Prices are eroding, prices are going down for flash, but also for hard disk drives as we have ever-higher capacities for the same cost versus price.
As long as we as an industry manage to do that, we will have hard disk drives.
From what I know is happening in our laboratories, in our production and preparation for production, I can say that as of today, we could probably launch a 40TB hard disk drive, but this will cost several thousand dollars.
Why would someone buy a 40TB hard disk drive for several thousand dollars if they can buy two 24TB drives for several hundred dollars? Launching new technology for a higher price per TB wouldn’t work.
We have to launch new technologies for the same price per TB, and it’s pretty clear that for 24, 28, 30-something – 40, probably up to 50 – we will be able to do that. Already, the technologies for that have finished their research phase. They are now in production. We have to bring the production cost down. We have to establish the economics around it. So, we can say that for the next, say, 10 years, hard disk drive is safe.
The other thing that is technically possible would be a hard disk drive of 100TB to 200TB. That’s confirmed by research. So, 100TB to 200TB would be physically possible in the current form factor.
What we don’t know yet is whether we can manufacture it for the price the market will accept. If we manufacture it for the same price as an SSD, it will be too high. There needs to be the gap, and hard disk drives may slowly phase out at some point in the future, when we cannot manage to follow that trend, say, higher capacity or the same cost.
But it’s not yet known if this will ever happen. And also, there are guys out there who say [for] flash technology, due to its speed, you can do much more compression and deduplication; other features.
And also, some people say flash, although it’s a higher capex initially, [by a] factor [of] seven, you will save on opex.
Well, at the moment, we are at factor seven. There’s a long way to come to a factor of 1.5 to factor 2.
And also, [there’s] the argument with the saving on opex. We have done some calculations. With a capex error of factor seven, the opex savings in terms of datacentre space are much smaller. [There are] savings on energy consumption and cooling – SSD-based storage consumes a little bit less energy and can be more efficiently cooled – [but this] amortises in 15 to 50 years, depending on which prices you assume. Let’s say 20, 25 years. So, that’s a no-brainer today.
You wouldn’t save with any type of flash against HDD, unless you can use its high performance. This implies smaller storage, and this is what people are doing today.
The larger [deployments] from all arguments will stay for a long time on spinning disks.
