Talk to most people about servers and their eyes start to glaze over. After all, if you’re not an IT professional, it’s not exactly a great dinner party conversation.
The truth is, in the era of cloud-driven applications in which we now live, servers play an incredibly vital role, functioning as the invisible computing backbone for the services upon which we’ve become so dependent.
Most servers live either in large cloud-hosting sites or within the walls of corporate data centers. The vast majority of them are Intel x86-based computing devices that are built similarly to and essentially function like large, powerful PCs. But that’s about to change.
Given the tremendous growth in the burgeoning world of edge computing—where computing resources are being pushed out towards the edge of the network and closer to us and our devices—we’re on the cusp of some dramatic changes in the world of servers. The variations are likely to come in the size, shape, capabilities, number, and computing architecture of a whole new category of devices that some have started to call gateways or, in more powerful forms, edge servers.
The basic idea driving edge computing is that current centralized cloud computing architectures are simply not efficient enough for, nor capable of, meeting the demands that we will soon be placing on them. Thanks to new types of applications—everything from voice-based personal assistants that use the cloud for translation, to increasingly connected cars that use the cloud for mapping and other autonomous features—as well as the continued growth of existing applications, such as streaming media, there’s an increasing recognition that new types of computing infrastructure are necessary. Distributing more computing intelligence out to the edge can reduce latencies and other delays, improve network efficiencies, reduce costs, enhance privacy, and improve overall capacity and performance for intelligent services and the connected devices which rely on them.
Because this intelligence is going to be needed in so many places, for so many devices, the opportunity for edge servers will be tremendous. In some instances, these edge servers may end up being downsized versions of existing servers, with similar architectures, similar applications, and similar types of nearby connected infrastructure components, such as storage and networking.
In many more cases, however, edge computing applications are likely going to demand a different type of server—at many levels. One likely scenario is best exemplified by hyperconverged server appliances, which essentially provide the equivalent of a complete data center in a single box, offering intelligent software-controlled storage and networking components, in addition to the critical compute pieces. The beauty of hyperconverged devices is that they require significantly less space and power than traditional servers, but their software-based architectures make them just as flexible as large data centers. This will be critical for edge servers because the need to have them be reconfigured on the fly to meet rapidly shifting application demands will be essential.
Another likely scenario is a shift towards other types of computing architectures. While Intel-based x86 dominates the very conservative traditional server market, the fresh approach that edge-based applications servers and applications are likely to take removes the onus of legacy support. This will free companies to choose the types of architectures best suited to these new applications. A clear potential winner here is Arm, whose power-efficient designs could find a whole new set of opportunities in cracking the server market for edge-based devices. A number of vendors, including HPE, Cavium and others are just starting to deploy Arm-based servers and edge computing applications will likely be a strong new market for these products.
Even within x86, we’ll likely see variations. With AMD’s well received Epyc line of server chips, there will likely be more acceptance of it in edge server applications. In addition, because many edge computing applications are going to be connected with IoT (Internet of Things) devices, new types of data and new types of analytics applications are going to become increasingly important. A lot of these new applications will also be strong users of machine learning and artificial intelligence. Nvidia has built a strong business in providing GPUs to traditional servers for these kinds of AI and machine applications already and they’ll likely see even more use in edge servers.
On top of GPUs, we’ll likely see the introduction of other types of new architectures in these new, edge servers. Because they’re different types of servers, running new types of applications, they’re the perfect place for vendors to integrate other types of chip architectures, such as the AI-specific chips that Intel’s Nervana group is working on, as well as a host of others.
Software integration is also going to be critical for these new edge servers, as some companies will opt to transition existing cloud-based applications to these new edge servers, some will build tools that serve as feeders into cloud-based applications, and some will build new applications entirely, taking advantage of the new chip architectures that many of these new servers will contain. This is where companies like IBM have an opportunity to leverage much of their existing cloud and IoT work into products and services for companies who want to optimize their applications for the edge.
Though most of us may never physically see it, or even notice it, we are entering a phase of major disruption for servers. The degree of impact that edge-focused servers will ultimately have is hard to predict, but question of whether that impact will be real is now a foregone conclusion.