IBM 3850, the first ever robotic tape library (1974). [A black and white photo of wall storage shaped like a honeycomb. The library could access up to 16 tape reels simultaneously.]
This piece originally ran in the Are.na Annual vol. 6, themed “document.”
In 2011, the same year that Google launched their ill-fated social network Google+, the company lost the email data of tens of thousands of customers. As a software bug spread through Google’s servers, Gmail users logged in to see empty inboxes. The data was not just misplaced, it was deleted, along with all of its copies. The unexpected hero of this story turned out to be magnetic tape cartridges, a medium that first gained popularity in the 1950s.
Since magnetic tape storage uses different hardware and software, it wasn’t subject to the same bug as the hard drives, and the emails could be safely restored. Google’s biggest consumer services — Gmail, Drive, and Photos — all plan ahead for data disasters like this one by backing up their storage on magnetic tape cartridges. Magnetic tape works roughly the same as the paper punch cards developed by Ada Lovelace in 1843, except that instead of holes in paper representing binary code, magnetic polarities (North and South poles) are imprinted on magnetized plastic tape reels. When magnetic tape was introduced in the ’50s, it had greater durability and storage capacity than paper punched tape; since then, we’ve seen storage technology advance to magnetic-core memory, hard drives, flash memory, solid state drives, and, finally, cloud computing — the pinnacle of modern data storage. But behind the scenes, magnetic tape has remained the unlikely backbone of cloud providers’ archival practices.
Univac advertised its metal reels as more durable and efficient than punched paper reels or cards. [A black and white advertisement from the ’50s for magnetic tape, “saves 90% in storage and handling over punched cards.”]
How did magnetic tape come to occupy this position, as the final physical backup for our data storage? I grew up in the 1990s playing cassette tapes on my Sony Walkman. I know first hand how easy it is to destroy a tape. Fingerprints, dust, and scratches caused glitches during playback. Sometimes rewinding, pausing, and fast-forwarding triggered the player to eat your tape. The slightest touch when removing a tape from its case could loosen the reel and cause an extra piece to hang out. To fix this problem, you’d have to manually advance the second spool with your pinky finger or a pencil, until the loose tape was stretched taught again. This fragility hardly inspires confidence as the technology powering the cloud, but large companies like Google, Microsoft, and Amazon must be using it for a reason.
[A black cassette with a white, minimalist label.]
Every time we upload a file to a server, it takes up space. “Hot data,” or data that’s accessed often, is stored on Solid 98 State Drives (SSDs), which have the fastest read time. This essay, which is being written in a Google Doc, would be considered hot data, since the file is updated each time I edit the words on the page. As data “cools down” and is accessed less often, it can be ported to Hard Disk Drives (HDDs) and eventually, magnetic tape cartridges. At the time of this writing, magnetic tape is still the cheapest long term archival “cold” storage option. Back in the 1950s, computers were primarily used by banks, insurance companies, and the US Census Bureau to run mathematical computations on datasets. Each reel of tape stored at most a few megabytes of data, the equivalent of a single JPEG photo. Today, we have closed cartridges instead of open reels, which means there’s no danger of damaging the film by touching it. The latest generation is called LTO-9 and can store 45 Terabytes (TB) of data per cartridge. That’s 200,000 iPhone photos, at an estimated 5 Megabytes each. Considering how many people use Google Photos for backup, and how much storage that requires, it’s clear why cloud providers are looking for the cheapest possible solution. Purchasing a single Fuji LTO Ultrium tape at retail prices costs $84.95, which works out to $1.89 per 1TB. This is much cheaper than other mediums like flash or optical storage, and it’s on track to get even cheaper. According to a roadmap released by the LTO Program Technology Provider Companies, each new generation of LTO tape will double storage capacity, with LTO-10 estimated at 90TB per cartridge.
“LTO” stands for Linear Tape Open format, which references the linear nature of a tape reel. The pickup head doesn’t move; rather, the tape film moves back and forth so you can find the data you seek. It’s just like rewinding a VHS tape so you can watch a movie scene again. Because of the time needed to position the tape, accessing LTO data takes longer than flash storage like SSDs. So while LTO tape is cheap, it’s also slow, and it lasts no more than 15-30 years, requiring frequent replacement. If the holy grail of data storage is a medium that’s fast, cheap, and long-lasting, that’s yet to be realized.
Spectra logic LTO ultrium 10 pack. [A sleeve of 10 square tapes.]
But researchers are trying: Today, there are attempts to store data on DNA, for example, a medium that would last much longer than tape and have a greater possibility of being decoded by future civilizations. Meanwhile, with Project Silica, Microsoft is researching a long-term alternative that would write data on silica glass plates. Deputy lab director Ant Rowstron has said that, “primitive peoples had better success with cave paintings and hieroglyphics than we have storing content for the future.”
**
Within my lifetime, I’ve seen data formats change so many times I can barely keep up. The amount of information we’re creating and storing on the cloud is massive. The last time someone tried to measure exactly how massive was in 2011, with an estimated 23 Exabytes (EB) of information transmitted online every seven days. That’s 4.6 trillion iPhone photos every seven days. Consider that number alongside another: there are currently 8.2 billion people on earth.
In 2016, Cisco Systems explained that “the largest contributing factor to the growth of IP traffic comes from video traffic (including online streaming services like Netflix and YouTube).”1 The more data we create and transfer, the more data centers we’ll need. Each data center runs on electricity and water for cooling systems, and requires people’s atten- tion for maintenance and security. Google alone currently has 25 data centers worldwide. This figure doesn’t account for enterprise Google Cloud regions and zones, so the true number is likely much greater. Eventually, all the hot data we create is going to cool down, and when that happens, cloud providers will have an enormous “cold” storage problem.
Courtesy Library of Congress. [One man looks on as another man prepares Univac computer to predict a winning horse.]
In order to store so much data, companies have built tape libraries with robotic arms that find and file cartridges. In 2011, Google confirmed they had 16 Oracle StreamLineTM SL8500 tape libraries, each of which can store 800 LTO cartridges in its robotic interface. Inside any cloud data center, these large libraries serve as a way to quickly locate and transfer data from cartridges to another medium, like a hard drive. But it’s not as simple as just playing back the “cold” data on the tapes. For security reasons, your data never resides on the same drive. Instead, it is broken out into small chunks and encrypted. When you upload a document to the cloud, it’s broken up into small pieces and transferred to multiple drives worldwide, like ghosts in the machine.
Google’s policy is to “randomly name these data chunks as an extra measure of security, making them unreadable to the human eye.” That way, even if a hacker were able to access a drive, they wouldn’t know how to read its contents unless they also had access to Google’s proprietary software. Here, a document becomes more than just a file — it’s also the software needed to decrypt the distributed pieces of the file, then load it into an interface like Google Docs. “As systems get built on top of cloud providers, or platforms like Facebook, storing them would [also] require recording all that backend software that is not shared.”2
[A black room with dozens of tapes stored on shelves along the wall, a black robotic arm in the aisle between them.]
In our current manifestation of the internet, a document is no longer confined to just one file or server, and no singular piece of software can read it. You can’t save a Google Sheet to your computer. Instead, you must export it to a Comma Separated Values (csv) file, losing version history and proprietary Google functions. On the cloud, the line between a document and software are blurred to the point of near-invisibility. We must look closer and focus, so as not to lose track of the distinction completely.
**
Growing up in the ’90s, I remember the death of sites like Geocities, Angelfire, and Xanga, and all the content we collectively lost. I remember the shift from movies on VHS tape to DVD to BluRay to streaming — the high cost of adapting to each update, and the hardware that filled landfills after becoming obsolete. Even in 2019, Myspace lost over 50 million songs uploaded from 2003–2015 during a server migration gone wrong.3
In 2022, Dawna Kerney, a resident of Fresno, CA, bought what she believed were blank VHS tapes from a thrift store, but found that they contained footage of a couple and a newborn baby. Instead of recording over it, she searched for the original owners of the video, going as far as contacting the VHS conversion company whose name was in the footage. She told her local news station, “I just thought this belonged to somebody and it appears to be possibly from the ’50s.”
Courtesy Fox News KMPH. [A close-up of Dawna Kerney’s hands holding the VHS tape she found.]
When I imagine a story like this unfolding at a cloud provider like Amazon Web Services, I can’t see them treating a found document with nearly as much care. They have too many files they need to keep online; identifying the owner of a wayward home movie isn’t scalable. For our most precious memories, we should seek longevity as a North Star, and we should hold our data close. Think of a wallet with a fold out section full of photographs of loved ones—what data should we keep in our back pockets?
This is not to suggest that everyone should go off-cloud. I’ve tried to “de-Google” my life many times, but for me it comes at too high a price — losing the ability to collaborate with peers, schedule calendar events with friends, or even access documents. Instead, I’d like to propose that we carve out a new space for precious documents we would like to preserve even if the cloud wasn’t around. Do the curatorial work to keep our data fresh. Export your Are.na channels. Take screen recordings of websites that you like, to supplement the work of the Wayback Machine. Prune your harddrives from time to time; convert files to long-lasting formats like JPEG, PDF, or JSON. Capture and archive the interfaces, thoughts, and documents created in software systems that are always changing.
There’s an aphorism that once you put something online it’s there forever, but this is proving to be less and less the case. And as we see with magnetic tape storage, the cloud is very much a material thing—backed up by hardware and maintained in sprawling facilities, with a real and growing environmental cost. These conditions call for us to be more discerning about the caliber of memory we record. Instead of acting like a cloud storage provider and putting everything in “cold” storage, let’s keep our most treasured memories warm.
[1] “Cisco Annual Internet Report,” Cisco, 2020.
[2] Sebastian Moss, “The Last Data Center,” Data Center Dynamics, 2022.
[3] Roisin O’Connor, “Myspace loses ‘over 50 million songs’ from website after server migration project goes wrong,” The Independent, 2019.
Megumi Tanaka is a designer and educator interested in building gentle web systems.