The desire for more agility, the move to open systems and virtualization, and the need to scale more efficiently are changing networking as we know it.
That’s impacting everything from customer premises equipment; to network elements like application servers, firewalls, routers, and switches; to the systems that manage the devices and bill for services; right down to the boxes and racks that contain the equipment.
Facebook (News - Alert) is in large part responsible for these changes. As we discussed in our May cover story, the social networking giant helped get the Open Compute Project off the ground. Several years ago Facebook, along with Andy Bechtolsheim, Goldman Sachs, Intel, and Rackspace, launched OCP in an effort to help spur the creation of hyperscale networking solutions, which weren’t available in the marketplace at the time.
“Facebook’s Open Compute Project is making server hardware an open
source option,” John Day, vice president of sales and marketing at Anord Critical Power Inc., explained in the first quarter issue of TMC’s (News - Alert) Cloud Computing magazine. “Now companies building data center facilities can use, modify, and optimize their designs – all for free.”
Since Facebook launched OCP in 2011, he added, the company has saved $2
billion. And Fidelity Investments, which is among the other companies that have employed OCP, Day said, has saved 20 percent on its data center electric bills as result.
Recognizing the benefits of OCP, communications services providers like AT&T, DT, Orange, SK Telecom, and Verizon then wanted to get involved with OCP. And then many in the supplier community, like Cisco, HPE, Huawei, IBM (News - Alert), Microsoft, and Nokia, did too.
Bill Carter, the CTO of OCP and a 33-year veteran of Intel, said it became clear a few years ago that telcos were interested in looking at non-proprietary solutions. Getting them there involves moving them from purpose-built appliances to software packages hosted on mainstream enterprise computers. That’s the NFV transformation the telcos are in the midst of now, he said. It also entails figuring out how to enable service providers to obtain true openness after they’ve been through the first phase. That, he said, led to the creation of the OCP’s Telco Project.
OCP’s Telco Project is looking at how carriers can leverage open hardware platforms and technology in hyperscale cloud data centers and the central office side of their businesses, Carter explained. Indeed, the telephone company central offices of the past are going to look a whole lot like data centers in the future.
Verizon is among several carriers that have recently been working with Radisys on a rack architecture that leverages new thinking about power and shared access. That involves using standard-sized 19-inch racks, but offering the ability to plug into them from the front, while providing optical and power connections on the back. In this architecture, all the power comes from a shared power source, Carter added, so it’s efficient both from AC to DC and phased balance perspectives.
That was inspired by the Open Rack architecture that came from Facebook. But the Radisys solution is based on a new design called CG-OpenRack-19, with the CG standing for carrier grade. The CG-OpenRack-19 design (to which Pentair also contributed) was contributed to the OCP in December.
OCP supports four types of cabinets, Carter explained. There’s Open Rack; a recent revision to OpenRack, which features a 48V DC option; CG-OpenRack-19; and Olympus, which Microsoft (News - Alert) contributed.
It’s good to have a few choices, Carter said, because every company has different requirements on this front based on their varying business continuity, climate control, and other needs and preferences.
Open Rack is based on a 600x1200mm layout but features an improved rack design, without a lot of space on the sides, Carter explained. It made the IT bay wider (21 inches as opposed to 19 inches), which allows for three (as opposed to just two) compute nodes. Open Rack also has bus bars around the back that feed off of a power shelf. And the power shelf allows for three phase input, which allows for phase balancing.
“So that’s a huge efficiency opportunity,” said Carter.
Open Rack also has a slightly taller rack unit, which has better air flow and cooling, he said. That means users need less climate control, and in some cases can eliminate chillers completely.
As for the recent 48V revision to Open Rack, that allows for higher powered racks to get the highest efficiency conversion (with a 1 to 2 percent improvement in losses). Google (News - Alert), which recently joined the OCP, was involved in bringing this option to the OCP.
Meanwhile, Olympus distributes AC power to the servers. So it allows users to connect to dozens of different AC power distribution schemes. It also works with legacy 19-inch environments, Carter said, so although you will need a new rack with Olympus “you can shove your old equipment into it.”
Of course, OCP is just one of the industry consortiums assisting the telcos, and even cablecos like Comcast Cable, in transitioning to more data center-like infrastructure. Open Networking Lab (or ON.Lab, which is the entity responsible or the Open Network Operating System) and the Open Networking Foundation are a couple of others. These two entities, which plan to merge in the third quarter, have been working on something called CORD.
ON.Lab and The Linux Foundation introduced the CORD concept a while back. And last summer CORD become a project within The Linux Foundation, which is home to such other open source projects as OpenDaylight and ONOS.
CORD stands for Central Office Re-architected as a Data Center. This concept combines the cloud, commodity infrastructure, NFV, open building blocks, and SDN to bring the agility of the cloud and the economies of scale found in the data center to service provider networks. That spans from the equipment at the home or office customer premises, to the access part of the network, to the telco’s central office.
(The OCP discussed earlier in this article does not have a formal relationship with CORD, Carter said, but OCP and CORD are complementary. CORD is focused on rearchitecting telco facilities to be more like data centers, he said, and they’re doing that with cloud-class hardware.)
Interestingly, Google hosted the first CORD Summit last summer at its Sunnyvale, Calif., headquarters. Not surprisingly, many of the world’s largest telcos are involved with CORD as well.
For example, in August there were reports that AT&T was conducting a field trial of CORD in Georgia. And in a recent interview with INTERNET TELEPHONY, Patricia Chang, senior manager of technology and architecture at Verizon, said Verizon is working with AT&T, SK Telecom, Intel, and Radisys on a mobile version of CORD known as M-CORD. That’s attractive to Verizon, she added, because of its openness and time-to-market benefits.
M-CORD aims to lay the foundation for 5G cellular networking and services through support for disaggregated and virtualized evolved packet core, end-to-end slicing from RAN to EPC, mobile edge computing, and programmable radio access network.
Slicing will come in handy for Internet of Things applications in which the IoT device needs just a small slice of the network, explained Chang. It also makes sense for broadcast events that require network resources only for a limited period of time, she added.
“The open source driven model is a fundamental element for the 5G roadmap as it enables extraordinary agility for identifying and responding to subscribers’ needs far more quickly than traditional mobility standards,” said Joseph Sulistyo, senior director of open networking solutions and strategy at Radisys Corp. “The integration of open source EPC enhances the M-CORD platform to provide boundless core service flexibility and brings us a step closer to unlocking the promise of 5G.”
M-CORD capabilities were demonstrated at the recent Mobile World Congress. But Chang wouldn’t speculate on when M-CORD solutions might be ready for implementation.
Edited by Alicia Young