When Verizon announced last October it would be entering the enterprise-grade public cloud market it was with the promise of providing an object storage service so reliable and so elastic that movie and TV studios could eventually come to trust it for storing and securing all their media. But Verizon was not ready to explain exactly how that object storage would integrate with its Terremark data centers, until yesterday.

In a joint announcement at Cloud Expo East in New York City, Verizon introduced its object storage partner for public cloud services: Amplidata, which unveiled a new object storage architecture called Himalaya. In an interview with FOCUS, Amplidata CEO Mike Wall said Himalaya will be unique, in that it’s scalable in three dimensions.

“What we’re talking about is something that no one else in the industry has: three dimensional elastic scalability,” Wall says. “What this architecture enables us to do is scale capacity, scale performance and scale a single namespace, and within that namespace an unlimited number of objects — trillions of objects in a single namespace — independently.”

Wall says when Verizon was scouting for an object storage partner its key criterion was that the partner’s scheme utilize a scalable namespace, independently of any other property. Knowing Verizon’s history, one might come to the conclusion that this would enable it to concoct ‘good/better/best’ service tiers, and charge premiums. That’s not the objective, according to Wall.

“I remember the meeting with Verizon in early 2013,” he says.  “They said, ‘Look, we want a system where if we had the largest company in the world with a namespace, we do not want its users to sign onto different namespaces to access different data.’”

Himalaya explained
Wall explains that Himalaya will enable clients to have individual user buckets whose capacity is essentially limitless. But it will enable the service provider to present this bucket as the entirety of the namespace belonging to the user. In other object storage systems, such as Amazon’s S3, buckets are essentially interfaces to separate namespaces protected by individual keys. While Himalaya’s buckets will be interoperable with S3 at time of release, in Himalaya they contain all the data — including from other data providers to which the client may subscribe — to which the user is entitled. (Amplidata promises compatibility with OpenStack’s Swift object store later this year.)


The Amplidata Himalaya architecture

Typical object storage architecture pairs the metadata that describes objects with the objects themselves. It’s said that the upshot of this is to ensure that each object is self-contained and integral. Himalaya takes a gamble that thus-far unlimited scalability can be achieved by physically attaching the metadata to an entirely new physical layer of nodes.

To explain:  Amplidata’s previous object store architecture, called AmpliStor — which, by the way, formed the basis of Intel’s Object Storage Reference Design (PDF) — expedited access to data objects by caching their metadata in the controller nodes.  This way, objects that have already been fetched into the controller cache need not have been reacquired from the broader storage pool.

As Wall explains, Himalaya replaces AmpliStor’s scheme with the creation of a second tier of nodes above the controllers, which Amplidata calls the scaler level (no, that’s spelled correctly). Reverse proxy servers are also deployed at this level, to provide protected access to the storage pool.

“If you think about how EqualLogic and LeftHand Networks used to talk about scale-out storage,” the CEO continues, “they would take a 2U iSCSI-based storage array and scale it out... to a 10 or 20 TB system. Think about the base unit of Himalaya being three to six racks — 12 to 20 petabytes using today’s capacity. That then folds into the scaler level, and that’s what we scale out with. So we can go to zettabytes with this system.”

Amplidata will be offering Himalaya as a product unto itself, including a special edition for service providers. This SP edition is geared to be readily integrated into legacy environments, Wall says, including on the service providers’ choices of hardware.

Verizon and Himalaya
It’s the SP edition, essentially, that Verizon will be incorporating. Last October, Verizon opted to deploy AMD’s SeaMicro servers in its Terremark datacenters.  SeaMicro servers may have Opteron processors, and they may have Intel Xeons.  Over the past year, Wall tells FOCUS, Amplidata has been working with Verizon to integrate Himalaya into its public cloud architecture, including its existing storage service.

Some 80% of Verizon’s own customers, according to Wall, want a hybrid cloud object store solution. “They want a large capacity object storage system on-premises, but then they also want it to be tightly integrated with — I say ‘tethered’ to — a private and/or public cloud subsystem.”

A separate enterprise edition will be offered to general business customers, with the goal of full deployment in existing data centers on existing hardware infrastructure — including ordinary COTS components — in mere weeks. “We provide the GUI, manageability, usability, supportability and through our partnerships we offer different types of front-end gateways, letting customers choose an ODM hardware supplier,” Wall says.

The CEO also tells us to expect “waves” of Himalaya support-related announcements for thus-far unnamed partner vendors, in the coming months.