Cisco UCS B-Series Server Network Adapter Options Overview
Sept. 12th, 2010
Here’s a quick overview of the server network adapter options for Cisco’s UCS B-Series blade servers. Cisco has three classes of adapters:
- Cost Optimized:
82598KR-I (Intel), M61KR-I (Intel), and M51KR-B (Broadcom)
- Compatibility Optimized:
M71KR-E (Intel+Emulex), M71KR-Q (Intel+QLogic), M72KR-E (Emulex), and M72KR-Q (QLogic)
- Virtualization Optimized:
Here are the interfaces that an OS ‘sees’ depending on which adapter you choose (click to enlarge):
And the obligatory “Feature Comparison Chart”…
Some Frequently Asked Questions:
Q. Are there special FCoE drivers I have to load in the OS for any of the adapters that support FCoE in hardware?
A. No, the operating system is oblivious to the fact that FCoE is even being used – just like the external LAN and SAN is oblivious.
Q. Can I use FCoE in the 82598KR-I, M61KR-I, and M51KR-B?
A. Yes, you can use FCoE software initiator packages with certain operating systems. These software packages allow the use of a regular Ethernet NIC. Open-FCoE.org would be an example. However, you do lose many features of a hardware-based FCoE solution.
Q. Why would I need up to 58 interfaces (using Palo) in a single server?
A. Virtualization and flexibility. For virtualization, Palo provides this many interfaces so that our customers can use VMware Pass Through Switching (PTS) and Hypervisor Bypass (VM Direct Path). See this colleague’s blog for more details: http://www.unifiedcomputingblog.com/?p=116. For flexibility, Palo allows our customers to deploy a single network adapter that can be programmatically configured to provide the NICs + HBAs combination that any server in their data center needs. Instead of buying a different hardware configuration for different types of servers, a single hardware configuration can be purchased. If a file server needs 2 x NIC interfaces, a single Palo can do it. If a clustered server needs 2 x NIC interfaces for production, 1 x NIC interface for cluster interconnect, and 2 x HBA interfaces, a single Palo can do it. If a hypervisor host A needs 4 x NIC interfaces for production, 1 x NIC interface for Service Console, 1 x NIC interface for VMotion, and 2 x HBA interfaces, a single Palo can do it.
Q. Which NICs use which physical port on Palo? I see 56 NICs and two physical ports.
A. Cisco provides a feature called Fabric Failover, or fabric-based NIC Teaming, that allows all NICs to use a single physical port during a path failure condition on one side. During normal operation, the NICs are evenly distributed across the two physical ports. During a failure condition, all NICs can use a single physical port so that all NICs stay connected to the network. Fabric Failover happens in the fabric, as the name suggests, and does not require any OS driver or OS configuration. Fabric Failover is configured as part of the UCS Manager Service Profile for each server. The HBAs always use a single physical port and never failover – as any SAN admin would want it. Multipathing would be used instead.
Q. Where can I find information on supported firmware, drivers, Operating Systems, etc. for these adapters?
A. See the latest version of the UCS “Hardware and Software interoperability Matrix” here: http://www.cisco.com/en/US/products/ps10477/prod_technical_reference_list.html