Storage is evolving: it’s moving out of the data center into the cloud and, more recently, into edge. Edge is the English term used to designate all the production sites of a company that are not the headquarters. Until recently, it was all about installing NAS or out-of-the-box hyperconverged infrastructure in branch offices with no IT staff. Today it’s more about installing local storage to store surveillance video of shops, factory production data, connected equipment statements across a campus, city or vehicle.
To lighten the network load, this data is filtered and pre-analysed before being pushed to more advanced analytics engines in the cloud. Because, yes, businesses no longer just use cloud storage services for their only backups. Now they put the production data.
The challenge of simplifying storage for the Edge
Edge and cloud manufacturing carries the risk of dramatically increasing storage complexity. It starts from a data center where CIOs deploy massive SAN bays whose performance they configure infinitely, to a cloud where there’s no clock and production sites where there’s no space.
There is also the problem of the economic model. In the data center, we invest once and for all in equipment that we hope to make it viable for as long as possible. In the cloud, you pay every time you use something, even when it wasn’t planned and nobody budgeted for it. At Edge, we realize that the storage capacities needed are no longer proportional to the small team in a branch office. Feeds from a small local fleet of CCTV cameras, for example, require the same performance as document sharing at headquarters.
To solve these difficulties, suppliers must be inventive. And this is precisely the theme of this new issue of Storage.
In this thirty-third issue, let us first take stock of theedge storage. What exactly are we talking about? What technologies are used? Furthermore, the currently most documented solution to avoid problems would be to finally let artificial intelligence do the work.
In computer science, we speak by AIOps, or artificial intelligence at the service of IT operations. The principle is to run a Machine Learning engine for a certain amount of time so that it learns by itself the way computing works: who communicates what, with whom, how fast and how often? Then, the engine sends alerts when observed metrics start to deviate from the initial model. This principle was born in the field of cybersecurity, has extended to the operation of servers and, this interests us, now concerns storage.
Really innovative working products
That said, AIOps is an aid to diagnosing problems, but it is not the solution to avoid them. That’s why, after laying the groundwork for edge and AIOps, this issue of Storage looks at storage products that are truly innovative in how they work.
For example, storage solutions today benefit from absolutely extraordinary disks: NVMe SSDs. Because they’re made of Flash memory and plug into a PCIe bus as if they were extender cards, the NVMe SSDs are hundreds of times better compared to good old hard drives. We can condense a minimum of that into ancillary equipment, where it took whole racks of shelves and electrical inverters to put the hard drives into production. However, to take advantage of this performance, you still need servers and networks that support NVMe SSD speeds.
That’s where the innovations are. We are talking about the GRAID startup. He knows how to build a storage pool of 32 NVMe SSDs in RAID using a standard Nvidia graphics card, which costs less than 600 euros on the market. Usually it would have been necessary to assemble several controllers in parallel for more than ten thousand euros.
Again to save on the server side, ScaleFlux offers SSDs which in turn incorporate the processor cores. So obviously the initial idea of this startup was to do perform a variety of application functions in SSDs. For example, keeping only the images from a surveillance video where something is happening on the screen, cropping the images to the only license plates in a parking security system, etc. The challenge was to reduce the size of servers and networks to be put into a site.
But it turns out that the project was a little too cutting-edge: no company knows how to develop the functions that go into an SSD. As a result, ScaleFlux SSDs now only perform a turnkey function: they compress data on the fly. Get twice the capacity without having to use twice as many processors.
Liquid and Lightbits are reinventing the SAN architecture. The first proposes to replace shared storage bays with shared PCie expansion bays, in which we would insert both NVMe SSDs and accelerator cards. The interest is being able to compose much more modular installations, much more suited to the constraints of the production sites. It is also that the market will trend. Tomorrow, CXL technology, an extension of PCIe, will allow assembling clusters from much more functional modules simply connected to a network.
The second, Lightbits, is more interested in the economic component. Using Intel network cards, it becomes possible with its system to put SANs into production as fast as the fastest SANs in data centers, but from a simple Ethernet network.
Archive for even less
So much for the edge. As for data storage, the French Biomemory and the German PoINT are developing systems whose operational cost is ultimately much cheaper than the cloud. The downside of the cloud is that, even though inert storage is very cheap, each consultation of the data in the archives is billed at the maximum rate.
Biomemory’s solution is not ready yet, but it is promising: it consists in keeping the archive TBs in a powder capsule, more precisely on thebacterial DNA turned into specks of dust. PoINT instead inherits the long know-how of the people who invented the CD-R at Philips. The idea here is to store the archives in an object mode volume, which is made up of… the tape libraries. Combining the two techniques, PoINT manages to offer a solution as rich in metadata as the cloud and as durable as an on-premise safe.
A number of Storage full of ideas, on the eve of a great evolution of storage infrastructures!
Download the free file Memory number 33.