Remember the Sun Microsystems marketing tagline from the late '90s that proclaimed "The network is the computer"? Well, it's 2010 and that still isn't reality. But now it's at least close, and we're seeing very real features and products that suggest application awareness will be central to how networks function in the not-too-distant future.
Recent moves by Brocade, Cisco, Hewlett-Packard, IBM, VMware, and others underscore that switch vendors aren't content with supplying the hardware and software to make a dumb network for carrying data. They're demanding a bigger chunk of the computing market and putting up the R&D dollars to make it happen, investing in technology that lets the network morph into an intelligent, application-aware hub. If networking vendors have their way, the network of the future will control and manage the entire application stack--though they won't get control without a fight.
Superfast and low-latency connectivity is letting network gear vendors demand bigger roles. Other enabling technologies include virtualization and processors fast enough and memory abundant enough to allow abstraction of the hardware layer in networks, and eventually storage and other systems, just as it has been for servers. Big vendors are also promising that this "ecosystem" will be controllable from a single point, as well as able to monitor and proactively ensure its own health.
It's a realistic long-term vision, but one that must be approached with caution.
What's Driving This Bus
Server virtualization seeks to decouple compute resources from physical hardware--to make resources stateless. The network plays a vital role here. Before today's virtualization boom, the biggest challenge to distributing compute resources over any distance greater than the dimensions of a physical server was that no network could carry traffic at the speeds and distances required in a low-latency, high-speed configuration.
Old-fashioned buses could only take us so far. Now there are new networking technologies such as InfiniBand; Fibre Channel over Ethernet (FCoE); 10, 40, and 100 Gigabit Ethernet; N-Port ID virtualization; and lossless Ethernet. With these, IT can start to create a framework for true network convergence between Fibre Channel (the network for all things storage) and Ethernet (the network for all things IP). This convergence can bring a simplified data-center-wide "fabric" and allow seamless mobility of operating systems and apps among various compute resources.
Intelligent, application-centric monitoring and management tools are also critical to this vision. These tools allow for more proactive end-to-end monitoring of all compute resources in the environment and raise the awareness of the network in the context of the medium so necessary corrective actions can be made swiftly. These tools have become even more important in that they aren't external to the compute stack but can be embedded via hooks into the operating system, network, application, and I/O stacks. This improves their ability to be integrated with not only the compute side but also the network itself.
Naturally, switch vendors such as Brocade and Cisco, and management software vendors such as IBM and others, want to make the network not only aware of the high demand placed by such compute abstraction, but also put it in control of what gets accessed, and in what manner. When the network is conscious of what use these compute resources are put to, the idea is, the network can intelligently provide resources that are optimized for the task; for example, a CPU-intensive compute environment could be given resources that have faster, stronger CPUs vs. a memory-intensive compute environment getting resources that are optimized for high-speed memory access.
Furthermore, vendors are seeking to create a unified "ecosystem" where all compute resources are integrated and able to be controlled from a single point. In the app-aware network, it's the network that will have to monitor the health of the compute environment and the resources it uses and seek to self-govern and self-heal. If a certain application seems poised to fail or starts performing below set baselines due to a lack of adequate resources, whether CPU, memory, storage or throughput, the network will be able to automatically mobilize additional hardware in order to keep the system performing at an acceptable level.
On the "Internet-facing" or IP side, networks have become increasingly self-healing as well, thanks to intelligent monitoring and management software that can detect and isolate problems. In general, these systems probe the content being served from the compute resource. Expect this functionality to converge with server functionality one day.
An app-aware network will have to be almost robotic in the context of a data center: not only aware of what devices are plugged into it, but controlling when and how its resources are accessed. The network will "figure out" what the device needs, provide it with the proper resources, and when the time is right, disconnect it and/or shut it down.
Cisco's tie-up with EMC and VMware to form a "virtual computing environment" seeks to define this ecosystem. It also illustrates that no matter how big the switch vendor, making the network intelligent can't be a solo endeavor. EMC and VMware, as the storage and server virtualization leaders, will help direct what an app-aware network looks like.
Brocade has a similar vision. By acquiring Foundry Networks, the company has propelled itself into the IP and FCoE ecosystem and plans convergence of its Fibre Channel and IP lines. Similarly, Juniper is pursuing an aggressive IP strategy that is application aware and that provides greater and more granular control of end-to-end connectivity.
HP has products across the data center. It has networking gear from ProCurve and its recently acquired 3Com, which also brought it Chinese networking company H3C. It's generations ahead of Cisco in its servers and management software. HP and Dell are trying to counter Cisco's push into the broader data center by offering integrated blade platforms that feature converged network adapters and FCoE connectivity.
CIOs and data center architects need to make some decisions about how much they believe in this vision of an app-aware network. If you like the vision, hold off on large-scale investments in data center technology where possible until industry cooperation increases. Massive build-outs of data centers using Cisco's Nexus switches and unified computing systems, for example, are still unheard of, perhaps because of uncertain ROI amid high costs. Also, this is disruptive technology, and it's early in its deployment and therefore not yet stable--always a dangerous combination.
That said, we believe the app-aware data center will play a significant role in the future. CIOs will want to have architects examining how this technology could simplify their environments, where there's potential to make incremental steps toward this app-aware network, and where it will require more disruptive change. Only this analysis will make the cost footprint of implementation clear.
If you have not yet virtualized your data center environment as completely as you expect to, that road map is the best place to spend significant time and energy. Plan to upgrade from older versions of virtualization software, since many of the benefits such as dynamic resource allocation require newer versions. Staying all Fibre Channel SAN, IP SAN, or NAS at this point is prudent. Fix particular trouble spots by making incremental investments in newer, high-speed networks. For example, we expect FCoE to initially become popular as a "top of the rack" aggregation technology for simplifying data center cabling.
At the same time, examine your investments in management and monitoring software. Are your systems geared to take on the next wave, or are you looking at a massive rip-and-replace effort to handle new app-aware and even current virtualization technologies? Ask your major vendors how they plan to support app-aware networks, and whether supporting them means upheaval to their road maps.
In all this, the biggest risk at this early a stage is technology or vendor lock-in, which could cause expensive headaches later on. Stick to an ecosystem of best-of-breed vendors, and ensure they're in some level of harmony with industry approaches. Best not to pick vendors that are bitter rivals, but strive to balance the power each has.
The goal is simplicity, while maintaining choice and flexibility. Can an app-aware network truly commoditize computing, or will it end up overly complicating operations? Your choices will determine the answer, which makes it an exciting time to be running a data center.
Ashish Nadkarni is a consultant with GlassHouse.
Write to us at firstname.lastname@example.org.