Trending Topics
Trending
Return of the tape: the unlikely renaissance of tape storage and how IBM Research is leading the way
In a cramped, windowless room at the top of one of IBM Research Europe’s many buildings, tape is making a comeback. Not audio tape, not VHS, but tape as a storage and backup medium for data centres.
“A lot of people are kind of surprised that IBM still sells tape products, let alone does advanced research on the topic,” said Mark Lantz, who leads the team that is responsible for tape R&D at IBM Research Europe, Zurich. “But digital data tape as a computer technology has always been and continues to be important in data centres.”
In fact, Mark says, there has been “a kind of renaissance in the last couple of years” for tape. And that’s due to two trends.
Trends driving the return of tape in the data centre
The first is obvious: data storage demands in data centres continue to grow exponentially. “AI is probably the latest [driver]”, said Mark, who pointed out that it’s “very hungry for training data, but also produces lots of new data, new content”.
Another big trend is that hard disk drives are barely increasing their areal density year on year. Areal density being the amount of data that can be stored in a certain volume. Hard disks’ “historical average areal density scaled with about a 40% compound annual growth rate,” said Mark. “Over the last couple of years, it’s less than 3%.”
That’s important, because it means the cost of storage per gigabyte stays the same. And with data storage demands increasingly exponentially, the storage bill for data centres and hyperscalers also increases exponentially.
They need a different solution.
Cold storage versus hot storage
But here’s the good news: most of the stored data is hardly ever touched. “Sometimes, it’s never accessed, but there’s a motivation to keep it, sometimes for compliance reasons, or sometimes because, as technology changes, you may be able to use or monetise that data in the future,” said Mark. This is called cold storage.
Unlike hot storage, which needs to be available at all time with no latency, you can afford to have a delay with cold storage.
Traditionally, cold storage has appealed to financial services and scientific institutions, which need to store immense amount of data as cost-effectively as possible, but IBM is now seeing it grow in popularity outside these domains.
“The biggest growth in tape technology right now is amongst what we call hyperscale cloud companies,” said Mark. “Companies like Microsoft, Amazon – here in Europe, OVH, Baidu in China – that are either in the process of deploying tape or some of them using tape at an incredible scale.”
Environmental and security benefits of tape
Not only is the total cost of ownership “four to eight times lower” for tape versus hard disk, but because tape has no active power consumption it has a much lower carbon footprint. After all, it just sits there until you need it.
“How much lower depends on the details of the comparison,” said Mark. “So just to give you one concrete example, our latest enterprise tape library, what we call a Diamondback library, is a single rack with 27 petabytes of capacity. If you compare that to an OpenCompute 27 petabyte hard disk solution, the tape solution, has 96% less CO2 emissions over a ten-year lifetime.”
Don’t forget security either. “Tape has a built in physical air gap [because] the cartridge isn’t mounted in a drive,” said Mark. “You can’t modify the data. That gives you an extra layer of security.”
Future for tape storage in data centres and hyperscalers
But the whole reason for my visit to see Mark – part of a wider tour of IBM’s Research Europe labs – was the research he and his team is doing. And that, primarily, is to increase the areal density of tape.
“We have lots of potential to keep scaling the areal density and capacity of tape systems,” Mark explained. “Again, that may be surprising, since HDD [hard disk drives] and tape use the same basic magnetic recording principles, and so if HDD is slowing down, it’s kind of natural to think tape must also be slowing down. But it turns out, HDD has been much more aggressively scaled.”
That, he says, is purely down to resources. “[The hard disk industry] peaked at something like $40 billion a year in revenue and that enabled very large R&D budgets, which enabled the technology to be scaled very aggressively, so that today, a high-capacity 22TB disk has an areal density of something like 1,200 gigabits per square inch.”
The trouble is, that’s as high as you can go before hitting “some fundamental physical limits as to how far we can shrink the size of the bits that we write on the magnetic medium. The biggest challenge is caused by something called the superpower magnetic effect. And the HDD industry is kind of bumped up against that physical limit, so it’s getting very hard for them to keep scaling.”
But tape has much greater potential. “Today, our high capacity enterprise product operates with a 50 times lower areal density,” said Mark. “So from this fundamental physics point of view, we have lots of room to keep scaling the technology… until we will get to the same areal density where HDD is.”
And don’t forget that tape is removable. While a tape drive may be expensive, once you spread that cost over hundreds of cartridges the cost per terabyte becomes much, much lower than hard disks.
Talking latency
So cold storage has obvious advantages for cost, security and the environment. What about its main disadvantage: latency? Even in the best-case scenario, where you have a robotic system (think of a jukebox that holds hundreds of tapes) that will automatically fetch the right tape and insert it into the reader, it will take up to a minute to find the file.
The first two seconds of that are finding the right tape, explained Mark, with the rest finding the data stored on the tape. And bear in mind that a typical tape is around 1km long. This is why tape will always be far better suited to storing huge files, such as object stores or archived videos, than small ones.
Another historical problem has been ease of use: you needed to have specialist knowledge to use tape drive systems efficiently.
“One of the other things we’ve really been focusing on in the last few years is making tape much easier to use,” said Mark. “Our most recent product is what we call Diamondback S3 which is [a] single rack library that has a little top hat on top with a couple of servers and some SSDs as a buffer.
“We run a software stack on the servers that presents an S3 Glacier interface to the world. So you don’t need any specialised tape skills to run this. You can just buy it pre-loaded with media, roll it into your data centre, plug it into power and the network, and any data that you used to push to an S3 cloud, you can just push to this thing, and it costs much, much less.”
What’s next for IBM Research and tape?
You can already see the fruits of IBM’s research via its Diamondback family and its LTO products. “The latest LTO [LTO-9] has an 18-terabyte capacity,” said Mark. “Next generation of LTO 10 will be [around] 50 terabytes.”
It’s worth noting that LTO-10 is likely to land in 2025, four years after LTO-9, but Mark expects the gap between generations to narrow as demand increases.
“Our latest enterprise product has a native capacity of 50 terabytes, more than twice HDD, but lots of room to keep scaling the technology. That’s really one of the main focuses of my research team is developing the technology to enable the continued density and capacity scale.”
IBM works with media partners too. “In our most recent areal density demonstration, in collaboration with Fujifilm, we could show what we call single-channel recording of 317 gigabits per square inch, which would enable a single cartridge capacity of 580 terabytes. So more than half a petabyte in a single cartridge, more than ten exabytes in a single library.”
All of which means that the future for tape is much brighter than most people would imagine. You could say there’s a lot more in store.
Related news and features
About The Author
