Networks: The Crux of Tomorrow’s Datacenter


Source: LinuxWorld

SAN FRANCISCO — As enterprises virtualize their environments and move toward the next-generation datacenter, the network will enable this move, Rajiv Ramaswami, vice president and general manager of Cisco, said yesterday.

In his keynote speech yesterday at the Next Generation Data Center and LinuxWorld conferences, held concurrently at San Francisco’s Moscone Center this week through Thursday, Ramaswami told his audience that the network will have increased bandwidth and capabilities to meet the needs of virtualization.

He predicted that Ethernet will win out over other networking technologies, and that enterprises will consolidate their networks to meet their evolving datacenter requirements.

Ramaswami’s remarks are a continuation of Cisco’s Data Center 3.0 strategy laid out in July 2007 at the company’s Networkers conference in Anaheim, Calif. Since then, Cisco has unveiled several products in line with this vision.

The network is “the key to enable the other forms of virtualization,” because as enterprises virtualize more of their IT infrastructure, they will have to evolve to service orchestration, where “you pool every resource — your network, storage, your servers,” Ramaswami said.

This pooling will let an enterprise allocate resources as needed, improving datacenter utilization rates, Ramaswami said. These rates currently are only about 20 percent, he added.

As they move toward the next-generation datacenter, enterprises will continue to virtualize, and this will require more bandwidth in networks. “You’ll have multiple CPUs being deployed on top of which you run multiple virtual machines, and the net effect is you push a lot more I/O traffic into the network,” Ramaswami explained. (I/O stands for input/output.)

Networks will also require more bandwidth to handle the sheer number of connections. “If you have 1,000 physical machines each deploying 10 virtual machines, as far as the network’s concerned, you need 10,000 ports,” Ramaswami said.

This demand for more bandwidth will push enterprises toward replacing their current 1 Gigabit Ethernet networks in the datacenter with 10Gb Ethernet networks, Ramaswami predicted. Work on 40Gb and 100Gb Ethernet is under way, he added.

The IEEE has already decided that the next speed for Ethernet will be both 40Gb and 100Gb because they address different needs. The 40Gb rate will handle server and computing applications, and the 100Gb rate will handle aggregation and core networking applications.

Both standards are necessary because the areas they serve have different growth rates. Servers are doubling in their bandwidth needs every 24 months while core networking bandwidth needs double every 18 months, according to the IEEE.

Next page: The network of tomorrow

Page 2 of 2

The network of tomorrow

In addition to more bandwidth, networks will need new functions to enable scaling and mobility in the virtualized environment, Ramaswami said. Cisco is developing datacenter Ethernet networks that will let enterprises scale virtual machines (VMs) across datacenters using local area network (LAN) and storage area network (SAN) extensions, and will let them move VMs around the network, Ramaswami told his audience.

This project is in keeping with Cisco’s plans to be a player in the enterprise datacenter virtualization market.

The network of tomorrow will be able to identify VMs, conduct performance monitoring, automate and simplify the process of implementing security in a virtualized environment, and move policies associated with a VM along with it as it is relocated from one server to another, Ramaswami said.

Meanwhile, enterprises will begin consolidating their networks because having one network connecting all their resources together will provide efficiencies of scale, according to Ramaswami.

Datacenters have many different networks with several instances of each type, Ramaswami said. They have multiple Ethernet local-area networks (LANs), multiple storage area networks (SANs) using Fiber Channel, and high-performance computing infrastructure deployed over several instances of InfiniBand .

That means each server needs seven to nine adapters to attach to the different networks. With each adapter consuming about 20 watts of power, that adds up to quite a bit, Ramaswami said. A large datacenter may easily have 1,000 servers, he added.

Having a single network could “save you a huge amount of power and cost in the datacenter,” Ramaswami told his audience. It would also cost less to install because “you’ll be rolling out one cable instead of multiple cables,” he added.

Network consolidation would let enterprises to replace the seven to nine adapters each server uses with two 10 Gb Ethernet interfaces. Ethernet will win out over other networking technologies because “it has moved to the server motherboard from network interface cards (NICs), and is, effectively free,” Ramaswami said. “That hasn’t happened with the other networking technologies.”

None of this will happen overnight, Ramaswami said. The move to the next-generation datacenter and the new network “will be a five-year or 10-year journey,” he added.

News Around the Web