Cisco Eats Its Own Virtual Networking Dogfood
Cisco has been telling the world for about a year now to buy Cisco Unified Computing Systems (UCS) blades, integrate them with Cisco Nexus 1000V (N1KV) virtual switches and go virtual on a VMware platform. The company has decided to show the way to potential buyers (and set up a fancy demonstration site for them to boot) by building its own IT infrastructure around these components. There are already 4,000 VMs running on UCS, with a lot more to come.
"Across 50 data centers, we are about 50 percent virtualized," said Ken Schroeder, data center manager for Cisco's IT department. "We are migrating to UCS for all our virtual servers within one year."
UCS was Cisco's first advance into the server and blade space. It consists of a series of blades as well as a chassis which is designed to simplify deployment. It also harnesses virtualization to integrate server, network and storage management functions.
The Cisco UCS Blade Server Chassis includes unified fabric and fabric-extender technology to reduce the number of physical components. It doesn't need any additional management software and Cisco touts it as being more energy efficient than traditional blade-server chassis from vendors such as HP, IBM and Dell. This arrangement also cuts down on cabling. It makes use of traditional front-to-back cooling. This 6U unit can accommodate up to eight half-width, or four full-width UCS blades. Cisco brings a unified networking fabric into the chassis via up to two fabric extenders that pass all I/O traffic to parent fabric interconnects. This interface to the unified fabric reduces the number of adapters, cables, chassis-resident LAN and SAN switches, and upstream ports.
Cisco offers a range of different blades that fit within the chassis. Some of these blades are designed for the most demanding virtualization and large applications with a lot of memory packed inside. All servers utilize the latest Intel Xeon multi-core processors. Other blades have various features such as two dual-port mezzanine cards to boost I/O. Obviously the full width blades have more power and memory. But the half-width models are no slouches. They come with decent processors and lots of memory. This UCS hardware underpins the Cisco data center.
Prior to rolling out its new infrastructure, the growing usage of virtualization caused problems on the networking side.
"Our network team had no visibility into our virtual environment," said Schroeder. "It was difficult to troubleshoot before we implemented UCS and N1KV."
By deploying Cisco Nexus virtual switches, those issues went away. Reason: N1KV extends the reach of traditional network switches and the Cisco NX-OS right into the virtualization layer. N1KV places the Virtual Ethernet Modules (VEM) inside the hypervisor and manages them from external Virtual Supervisor Modules (VSM).
Aligning IT management with virtualization
Cisco also reorganized its IT management structure to align better with the virtual world.
A virtual infrastructure team manages VMware and UCS while the networking team manages the N1KVs and related network components. This has added in speed of service provisioning from Cisco IT to internal and external customers.
If a client needs a highly secure network, for instance, everything is automated based on the requirements. Networking staff can create the profile, enter it and it is automatically in place.
"In the past, we have had to deal with a whole lot of internal liaison, paperwork and running of wiring etc. before network profiles could take effect," said Schroeder.
Cisco gear is used elsewhere in the data center. At the access layer, Cisco UCS 6120XP 20-Port Fabric Interconnects are used to provide uniform access to both networks and storage. This 1U unit provides 20 fixed 10 Gigabit Ethernet and FCoE Small Form-Factor Pluggable Plus (SFP+) ports, 520 Gigabits per second (Gbps) of throughput, a single expansion module bay which can support up to eight Fibre Channel ports or up to six 10 Gigabit Ethernet ports using the SFP+ interfaces, and support for up to 160 servers or 20 chassis.
Cisco Nexus 7000 switches are utilized at the distribution layer. Designed for 10 Gigabit Ethernet networks, they enable a fabric architecture that scales beyond 15 terabits per second (Tbps). The fabric modules for the Nexus 7000 Series Chassis are separate fabric modules that provide parallel fabric channels to each I/O and supervisor module slot. Up to five simultaneously active fabric modules work together delivering up to 230 Gbps per slot. Cisco Catalyst 6000 series switches are used at the service layer.
"We have been able to put together a highly converged design with simplified management of our virtual servers, storage and networking," said Schroeder. "Our target is to achieve 80 percent virtualization or more."