How Did Cisco UCS Get to 10,000 Systems So Fast?

There has been a lot of press this week about Cisco UCS passing the 10,000 customer mark, which is an outstanding achievement. But I have found none yet that explains why it is getting so much traction. The basic reason is that when an organization uses more than two chassis worth of blade servers, the Cisco UCS solution costs less, is easier to manage, and has much less cabling, permitting better airflow for cooling. The system is able to do this because of a fundamentally different design than other blade servers.

Traditional blade servers have redundant power supplies and server blades in them. They also have system controllers, SAN cards, and LAN cards. These cards all connect the individual blades to the outside world, and since the blade server has to be reliable, the cards are usually redundant. That means each chassis has two controllers, two LAN, and two SAN cards.

Cisco did something fundamentally different with the UCS. They took the controller, LAN, and SAN connections and put them into a special box called a Fabric Interconnect. Each UCS chassis only has power supplies and server blades in it, and the dual Fabric Interconnects can connect anywhere from 8 to 16 chassis with good data throughput. That means that a company deploying Cisco UCS does not have to buy two controllers, two SAN cards, and two LAN cards for those 8 chassis, but just have to hook them up to the two Fabric Interconnects.

The Fabric Interconnects are the connection from the Unified Computing System to the rest of the datacenter. It has 10G ethernet ports to connect to the core LAN, and 8GB Fibre Channel ports that connect to the SAN. Everything outside the system is compatible with almost any brand of switch or storage. In many cases the organization also deploy Cisco Nexus switches, but it is not required.

Cisco designed other things into the Unified Computing System that makes it attractive for people that manage many racks of servers. All the fixed hardware information of the blades are programmable, so applications that are tied to MAC addresses or other specific features of a server can be migrated from one server to another, even if server virtualization is not being used. If the server is running a hypervisor, the network card can be set up to virtually split up the blade's dual 10G ports into multiple connections, which is necessary to properly operate a virtual environment.

The other major technical accomplishment that permits this revolutionary design is the full adoption of FCoE, or Fibre Channel over Ethernet, within the confines of the Unified Computing System. All the storage and network traffic from the blades to the Fabric Interconnects goes over FCoE, which is a combination of regular Ethernet and Fibre Channel traffic encapsulated within jumbo Ethernet frames. The storage traffic gets where it needs to go because Cisco uses Quality of Service mechanisms to prioritize the Fibre Channel traffic. At the Fabric Interconnects, the Ethernet and Fibre Channel are decoupled, and the storage traffic gets onto the Fibre Channel SAN just like it would from any other server.

There are actually storage manufacturers that are shipping FCoE interfaces on their storage arrays, allowing organizations to move away or not deploy a separate Fibre Channel SAN at all. This is the Unified Fabric that Cisco Nexus switches are optimized for, and permits even better performance and further cost savings.

Cisco has reached 10,000 customers for the UCS because they created a new design with proven technology that significantly lowers the cost deploying multiple blade server chassis. They also made the system easier to manage, and added lots of little details that add up to a product that is notably better than the chassis blade systems that have traditionally been deployed.

See my video whiteboard of how a Cisco UCS system works at Cisco UCS
There has been a lot of press this week about Cisco UCS passing the 10,000 customer mark, which is an outstanding achievement. But I have found none yet that explains why it is getting so much traction. The basic reason is that when an organization uses more than two chassis worth of blade servers, the Cisco UCS solution costs less, is easier to manage, and has much less cabling, permitting better airflow for cooling. The system is able to do this because of a fundamentally different design than other blade servers.

Traditional blade servers have redundant power supplies and server blades in them. They also have system controllers, SAN cards, and LAN cards. These cards all connect the individual blades to the outside world, and since the blade server has to be reliable, the cards are usually redundant. That means each chassis has two controllers, two LAN, and two SAN cards.

Cisco did something fundamentally different with the UCS. They took the controller, LAN, and SAN connections and put them into a special box called a Fabric Interconnect. Each UCS chassis only has power supplies and server blades in it, and the dual Fabric Interconnects can connect anywhere from 8 to 16 chassis with good data throughput. That means that a company deploying Cisco UCS does not have to buy two controllers, two SAN cards, and two LAN cards for those 8 chassis, but just have to hook them up to the two Fabric Interconnects.

The Fabric Interconnects are the connection from the Unified Computing System to the rest of the datacenter. It has 10G ethernet ports to connect to the core LAN, and 8GB Fibre Channel ports that connect to the SAN. Everything outside the system is compatible with almost any brand of switch or storage. In many cases the organization also deploy Cisco Nexus switches, but it is not required.

Cisco designed other things into the Unified Computing System that makes it attractive for people that manage many racks of servers. All the fixed hardware information of the blades are programmable, so applications that are tied to MAC addresses or other specific features of a server can be migrated from one server to another, even if server virtualization is not being used. If the server is running a hypervisor, the network card can be set up to virtually split up the blade's dual 10G ports into multiple connections, which is necessary to properly operate a virtual environment.

The other major technical accomplishment that permits this revolutionary design is the full adoption of FCoE, or Fibre Channel over Ethernet, within the confines of the Unified Computing System. All the storage and network traffic from the blades to the Fabric Interconnects goes over FCoE, which is a combination of regular Ethernet and Fibre Channel traffic encapsulated within jumbo Ethernet frames. The storage traffic gets where it needs to go because Cisco uses Quality of Service mechanisms to prioritize the Fibre Channel traffic. At the Fabric Interconnects, the Ethernet and Fibre Channel are decoupled, and the storage traffic gets onto the Fibre Channel SAN just like it would from any other server.

There are actually storage manufacturers that are shipping FCoE interfaces on their storage arrays, allowing organizations to move away or not deploy a separate Fibre Channel SAN at all. This is the Unified Fabric that Cisco Nexus switches are optimized for, and permits even better performance and further cost savings.

Cisco has reached 10,000 customers for the UCS because they created a new design with proven technology that significantly lowers the cost deploying multiple blade server chassis. They also made the system easier to manage, and added lots of little details that add up to a product that is notably better than the chassis blade systems that have traditionally been deployed.

See my video whiteboard of how a Cisco UCS system works at Cisco UCS