Mad Dog 21/21: If You Want Cheap Cloud Backup, Raise Your ARM
March 4, 2013 Hesh Wiener
If you are using a smartphone, tablet or PC, you can back up data to the cloud, 5 gigs or more, for free using Dropbox,Evernote, Google Drive, or Microsoft Skydrive. Want more capacity? It’s cheap. Want to share stuff among your machines or with others? Easy. Security? Pretty good even before you encrypt.
Compare this with IBM’s backup schemes for an IBM i shop. You could write a book about your experience: Fifty Shades of Blue. But do it fast, because disruption may be imminent, and you don’t want to get caught with your rants down.
Last month. Chinese search and cloud services biggie Baidu and American (if you overlook its Bermuda tax dodging headquarters) chip vendor Marvell, made a splash by showing off specially tailored ARM-based servers. The servers will initially be used for cloud data storage services, very likely the cold or warm rather than hot backup Baidu will provide for consumers. Generally speaking the users will upload lots of stuff, such as photos and videos and just let it sit more of the time. Once in a while users will download some of their files.
Conceptually, this is similar to what users do via Facebook, Microsoft’s cloud facilities, Google’s eclectic mix of web-based offerings, or the somewhat more focused cloud archiving technologies from Dropbox or Evernote. (For every one of these offerings that I have mentioned, there are plenty of alternatives. There are just too many to list here and now.)
One of the first, best and simplest schemes is the one from Dropbox, which basically lets you have a folder supporting a tree of subfolders and files that is just a click or two away on a computer, a tap or two away on a mobile device. Dropbox lets users designate two kinds of shared subfolders: some are openly available (for reading, anyway) while the others are tied to the Dropbox apps of other specific users. Your Dropbox can have many shared folders, each with its own sharing and permissions rules, plus totally private folders only you can see on your own client devices. All the other players in this segment offer similar or comparable capabilities but each one has its own idea of how to organize the access apps and how to provide end user support. There’s isn’t a best one, any more than there’s a best flavor of ice cream but some of these services are better known and more widely used than others.
All the outfits providing cloud data storage are naturally trying to run their server farms as cheaply as possible, which means finding the most cost-effective servers, software, and storage devices. Like all other data centers, the cloud storage server farms use a lot of electricity, making solutions that take less power and require less cooling more attractive.
With that in mind all the big names have tried off-the-shelf servers and customized ones. Some of the participants have gone farther than others, turning to ODMs (original design manufacturers) or the ODM operations of full range computer manufacturers such as Dell’s Data Center Solutions operation. Most of these services are using machines based on X86 chips, often from Intel. The operating systems on these machines are often based on Linux, but just how closely they resemble popular distros varies quite a bit. The disks used for storage are likely to be SATA drives, but more costly SAS (serial attached SCSI) drives are an option some of the players have tried.
When Baidu, a leader in the Chinese cloud world, revealed its plan to use servers based on Marvell ARM SoC processors along with other Marvell chips, machines that EE Times says will be manufactured by a Taiwanese company called Wywinn that no commercial end users in the USA had ever heard of, the story made quite a splash. The in-depth description published by The Register said that the machines were likely to be cheap to install, cool to run, and easy to stack and rack in large numbers.
The Baidu servers are 32-bit boxes (but with 40-bit addressing to accommodate larger main memory) that would never be leaders for computational or DBMS applications. But it turns out that the relatively modest amount of computing power required to manage disk storage. The machines are compact. A single 2U chassis holds six computing modules each with four 4 TB drives, for a total of 96 TB raw storage space. Most of the technical details remain obscured, but various published reports indicate the software will be a version of Linux-on-ARM, and TPM says it will be Ubuntu Server from Canonical.
What has happened here is that Marvell, which got its ARM development group by buying it from Intel, is trying to grab and hold some important market turf as its CPU engineers (and the circuit developers at dozens of other companies) figure out how to build a server-grade 64-bit ARM machine. The requirements are tough, as they are in the X86 world: Servers aimed for use in cloud applications have to be able to scale up to farms with hundreds of machines containing thousands of engines. But that’s what it takes to create a service that can run on the revenue generated by ads plus a relatively small collection of paying customers who are happy to buy more cloud capacity than their favorite service company will provide for free.
While all this is going on, IBM i shops are paying big money for backup technology that incorporates two IBM offerings, BRMS (Backup, Recovery and Media Services) for the IBM i hosts and Tivoli Storage Manager for clients and X86 servers. As you probably know, the Tivoli software ends up consuming its own server, and that server, wrapped around whatever data it has gathered from a user’s networks, can be sucked into the IBM i box. The user ends up with a complete data image but it’s a riddle wrapped in a mystery inside an enigma.
Drilling down to review some data that was originally stored by an app on a PC is just not very easy. By contrast, a home user or small business working with Dropbox or another cloud storage service can pretty much keep every file at the same level, the way the ancestors of the IBM i platform used to do things, or put lots of data objects in a relatively simple and easily searchable tree. Moreover, the cloud world gives users a lot more freedom than legacy systems when it comes to naming files. Photos can all have names starting “p h o t o” if that’s what one wants. Drawings might be the same file types (e.g., png) but can all start “d r a w” if one wishes.
It is pretty clear that home users and then teeny biz users have learned to work with cloud data storage. Large but still small businesses running systems based on Windows or Linux have an easy time hooking up with a cloud service, if they wish, and a growing number of offices have started training their users to keep all their individual work files and app data in folders on the cloud, sometimes in conjunction with local storage services, but increasingly not. The cloud storage schemes use incremental backup, so they stay pretty much up to date as events occur. Compared to local backup that has to be scheduled and which can become a big, slow, and annoying chore, cloud storage is a breeze. This article, for instance, was written with Word set up so its default file folder is subfolder in the cloud that is addressed more or less the same way a local folder on the C drive would be addressed.
(Actually, I store in both places using an app that keeps an eye on a designated C drive folder and its subfolders and copies changed data to the cloud in close to real time. If my computer poops out or my electricity goes away, hardly anything will get lost as I scramble over to a different computer or the backup battery-based laptop and wireless Wi-Fi hotspot I have and hope never to actually use. I have a few hours of PC power on tap, enough to cope, I think, as I contemplate buying a backup generator. And after that I can stay in touch via a smartphone that can run quite a long time. . . but which isn’t good enough for word processing no matter what the app vendors say.)
Anyway, Baidu wants to offer cloud services that cost less to operate than any of its home country rivals and less than the services offered by foreigners like Google and Microsoft, who would love to pick off the cream of Baidu’s user base, the relatively affluent and modern folk who can afford client devices and nice cameras that create a need for lots of storage. Baidu may also like the fact that systems built nearer to its facilities based on ARM intellectual property rather than Intel architecture might eventually garner prestige that it cannot get if its stands in line behind Google or Facebook at some big name company’s server factory.
Of course Intel and its clients don’t want ARM to take over a big slice of the server business. It is coming under a lot of pressure to dream up cheaper, faster, less power-hungry server chips and surrounding circuitry. And as the X86 world tries to cope with the disruption caused by Marvell and its ilk, IBM and other manufacturers will have to adjust, too. If storage subsystems based on ARM chips turn out to be superior for general purpose uses, not just for cool storage of rarely accessed archives, Big Blue’s storage subsystems, based on Power technology, are going to become uncompetitive despite their advanced software features. The same goes for EMC with its X86-based product lines.
So next time you are thinking about Option 21, you might also want to take a moment to examine alternatives that include cloud offerings. If Baidu, Google, Facebook and Microsoft are right, and it sure looks like they know what they are doing, cloud services are going to replace some and probably most of your backup software, disks, and tapes.