Deploying AI Models at the Edge

AWS articulates consistency as key to edge strategy, offers smaller appliances, more Local Zones

AWS articulates consistency as key to edge strategy, offers smaller appliances, more Local Zones

This is a guest post by Phil Shih, Managing Director and Founder of Structure Research.

AWS has again expanded its portfolio of infrastructure services aimed at addressing distributed and edge computing scenarios at its annual re:Invent conference. AWS made it clear that it wants to answer the question: what happens when AWS is not in a metro or location where the customer wants or needs to be? The answer is to bring the AWS experience to the customer no matter where they are. And that means new form factors and a wider range of deployment models that can work hand in hand.

Background and context: The move to support distributed infrastructure on AWS began just a few years ago. Organizations wanted to move from on-premise to cloud (and integrate the two), without having to move off the familiar VMware-based virtualization tools they had been using for so many years. Amazon obliged with the introduction of VMware Cloud.

Next came AWS Outposts, which is basically a converged infrastructure appliance that replicates the AWS cloud in a smaller, enterprise-friendly form factor. The appliance has the same API used on the AWS cloud and runs on the same hardware. It provides the same experience but connects back to a core AWS region for access to the full suite of AWS services. Customers can order one or literally dozens of Outposts racks (standing them up on-premise or in a colocation facility). AWS then delivers, installs and manages the gear for the customer.

After launching Outposts, AWS introduced Local Zone and then Wavelength. Local Zone is an AWS cloud infrastructure node that is significantly smaller than a standard AWS region/availability zone. From an infrastructure perspective, a Local Zone consists of a footprint within a colocation facility in an edge location (the first deployment was in Los Angeles). In the data center, AWS stands up the Outpost appliances. The idea is to get infrastructure closer to end users (where AWS does not have a data center/region nearby) and cut down the latency between them and the infrastructure. Finally, Wavelength was followed by the Local Zone offering. Wavelength is another deployment model built on Outposts. Instead of going into an edge location like Los Angeles with Local Zone, Outposts is rolled into a wireless aggregation site.

Decentralized and out to the edge: The idea behind all these new deployment models is to put AWS-based compute and storage out at the edge where AWS is unable, or more accurately, is not able to place hyperscale cloud infrastructure economically. This could be on-premise, in a strategic market like Los Angeles where certain target sectors are clustered or at the wireless edge to support performance-sensitive applications that are built with a certain degree of function separation. Getting closer means single-digit milliseconds of latency.

Outposts at the user edge: AWS is not done yet. It is now targeting user edge locations like restaurants, branch offices, wireless sites, retail stores or hospitals where there is either not much space or there are power and networking challenges. The solution is an even smaller Outposts form factor. Instead of a full 42U rack, AWS is making 1U and 2U Outposts servers. Importantly, once the server is connected back to the AWS network, end users will get access to all the AWS tools (just as is done with Outposts). AWS will remotely manage and update (patching, software upgrades, etc.) the Outposts servers just as it does with the AWS cloud. The larger goal is to create a smaller increment of compute that is integrated with AWS and still maintains a consistent user experience.

The rugged or disconnected edge: AWS is looking even further out to the edge of the enterprise. It has a number of different smaller Snowball-based devices for places like oil fields, farms and battlefields where supporting infrastructure may be unavailable. The idea is to collect data with the device and then transport back to connect with the cloud, either by physically moving the device or waiting until a network connection is available. There the user can take advantage of all the machine learning tools that AWS has.

Local Zone footprint update: AWS rolled out Outposts last year in Los Angeles and now has two nodes live and running. The next three Local Zone locations are going to be in Houston, Boston and Miami. An additional 12 locations will be rolled out in 2021: Atlanta, Chicago, Dallas, Denver, Kansas City, Las Vegas, Minneapolis, New York, Philadelphia, Phoenix, Portland, and Seattle.

Wavelength footprint update: Wavelength is in eight US metros. It is going global next year with KDDI in Japan, SK Telecom in South Korea and Vodaone in London.

Netflix uses Local Zone: Netflix confirmed it uses Local Zone deployments to help it deliver content to customers. Presumably, this is in Los Angeles.

Angle: The pieces have been coming together and AWS now has a much more complete edge computing story. Its strategy is clear. It will create more form factors so that compute infrastructure can fit in all the edge scenarios that are out there. A consistent user experience and API are maintained across any infrastructure form factor or deployment model and this will all tie back to its core cloud computing platform in centralized regions. The end game is clear: distributed, decentralized and integrated hybrid infrastructure.

Is this value chain going to involve other parts of the infrastructure ecosystem? It most certainly will. AWS can’t possibly deploy in every metro or in every building, and it surely can’t efficiently do it at scale. It will take advantage of existing infrastructure and that will translate into demand for third party colocation. That is good news for data center operators in edge markets outside the existing AWS footprint (also considering that the other clouds will drive further demand as they build out similar strategies). The smaller form factors should also create demand for capacity in micro data centers. We are seeing edge computing providers and CDNs put small increments of compute in micro data centers. Outposts’ servers could be going in there as well. Overall, expect AWS to continue pushing out to the edge and thinking of new ways of getting infrastructure to wherever it needs to be. The edge is clearly about extending and augmenting what is done at the core. The two are inextricably intertwined and what AWS is doing confirms that.

About the author:

Phil Shih is Managing Director and Founder of Structure Research, an independent research firm focused on the cloud, hosting and data center infrastructure service provider markets. (more)

DISCLAIMER: The views expressed in this contributed post are that of the author, and don’t necessarily reflect the views of EdgeIR.com. Contact us if you want to contribute a guest post.

Article Topics

 |   |   |   |   |   |   |   |   | 

Comments

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Featured Edge Computing Company

Edge Ecosystem Videos

Automating the Edge

“Automating

Deploying AI Models at the Edge

“Deploying

Latest News