The Visual Networking Index (VNI) is an ongoing initiative developed by Cisco to forecast Internet Protocol (IP) traffic over the next five years. The numbers are staggering. Global IP traffic is expected to enter the zettabyte range (yeah, I had to look it up, too). Much of the growth will be in mobile devices, with over 10 billion mobile devices and connections in operation by 2018. The VNI predicts a 61 percent compound annual growth rate (CAGR) in mobile, with video comprising 79 percent of all mobile traffic (Figure 1). That is a ten-fold increase in mobile traffic in just the next five years, with more than half of all traffic coming from non-PC devices. Bill Huang, Head of R&D at the China Mobile Research Institute, recently said, “China Mobile is driving to a 300x capacity increase with a 3x average revenue per customer increase.”
This landscape is creating major challenges for the telecom carriers, now frequently referred to as communications service providers (CSPs). They can’t provide ten times more throughput by just installing ten times more equipment and hope to stay in business. No one will be willing to pay $1,000 a month for their smartphone bill. New approaches are required. Providing functions that have traditionally been done with specialized hardware through software-based solutions like software-defined networking (SDN) and network functions virtualization (NFV) are getting a lot of attention and will help, but tomorrow’s telecom networks will require a lot more than rack after rack of commodity “pizza box” servers. Video and voice require lower latency and a much higher quality of service (QoS) than buying a book from Amazon. Specialized hardware and functions will still be required.
About a year ago, a group within PICMG began to conceptualize a next-generation, standards-based platform that could provide a ten-fold increase in processing power, storage, networking bandwidth, and cooling than what’s possible with AdvancedTCA (ATCA). That seemed like a good idea at the time, and many PICMG members who build equipment for the CSPs went out and talked to their customers about their needs. The feedback was that the industry could not wait five years for a big platform to be developed and tested, and that these customers want ways to increase network throughput quickly and more cheaply than before.
At the Open Server Summit, held in Santa Clara, CA in November, PICMG presented two papers on emerging requirements and held a well-attended round table discussion to kick around ideas. Chuck Byers, System Architect at Cisco, and Doug Sandy, CTO of both PICMG and Artesyn Embedded Computing, spoke about these requirements.
Mr. Byers pointed out that CSPs need more than simple compute servers can provide. Additional networking, storage, and various types of co-processing are required. Things like optical interconnects, graphics processors, and advanced cooling often don’t integrate easily into simple servers. Good security is becoming an important requirement and will be even more important in an “Internet of Things” world where big and sometimes dangerous equipment will be remotely controlled. A number of military users I have spoken with want security baked into the low-level hardware using mechanisms like hardware root of trust.
Doug Sandy echoed many of Chuck’s observations, and provided some of his own. With the help of Markinetics, a market research firm, Artesyn has dug into the issue of commodity servers versus specialized equipment for next-generation telecom networks. They believe that only about 10 percent of equipment needs can be provided by enterprise-grade commercial-off-the-shelf (eCOTS) hardware. A larger portion of their equipment must provide low latency, high availability, and be ruggedized to NEBS-level standards.
But CSPs also want to buy COTS gear from a wide range of suppliers, much like they do with ATCA. The “CSP COTS” piece of the pie is larger than the eCOTS portion, and should be an attractive market for many vendors (Figure 2). They want open standards. It will be up to organizations like PICMG to provide them, and to provide them as quickly as possible.
Doug also spoke to the trends in open-source hardware and software, pointing out that open source initiatives can develop solutions very quickly, but often require external oversight to manage distributions. Solutions based on open standards, on the other hand, tend to take a bit longer to develop but are generally stable for many years.
Both presentations are archived at www.openserversummit.com.