Cisco - UCS - Design Best Practices

1. Address Pools: best practices
1.1 UUID Pools

• UUID: global ID that is unique to a given server

• Composed of Prefix and Suffix

• Best practice: don‟t modify the prefix

• Recommendations:

• Use root “default” pool as the global default pool for all Service Profiles

• Populate the default pool with a block of 512 IDs

• Don‟t alter original Prefix, this is unique to this UCS

• Optional: choose a “Domain ID” for this UCS - used later in other ID pools

1.2 MAC Address Pools

• Prefer pools to burnt-in values whenever possible

• -Pools let you control the exact allocation following your naming convention

• -Makes it easy to identify a given blade or OS type on switches

• On the M81KR adapter (aka “Palo”) this is a must

• -M81KR can instantiate N interfaces  there are no burnt-in addresses

• Best practice

• -MAC pools: create pools that are multiple of 64

• -Optionally encode Domain/Site ID and OS Type

 1.3 World Wide Name Pools

• Using pools lets you communicate WWNs to SAN team ahead of deployment

• -pre-provision LUNs for boot-from-SAN

• -proactively perform zoning and LUN masking configuration

• One blade uses one Node WWN and as many Port WWN as there are vHBAs

• Node pool best practice: create one large pool that‟s a multiple of 16

• -Create the pool at the Root organization (you can use the default pool)

• -Zoning and masking does not use Node WWN

• Ensure node pools and port pools do not overlap

1.4 Port WWN Pools

• Always create pools that are multiple of 16 and contain less than 128 entries

• -This ensures vHBA0 (SAN A) and vHBA1 (SAN B) have the same low-order byte

• Counter-example using 233-entries pools

• Much better for both vHBAs to have the same low-order byte and a unique SAN Fabric identifier

• -Presence of “0A” or “0B” in the port WWN indicates SAN Fabric

2. Recommended Service Profile Designs

• Question: “how many vHBAs and vNICs should I assign to my Windows profile? How about ESXi?”

• -If you don‟t have a Cisco M81KR adapter, the answer is simple: 2 plus 2!

• -We‟re going to focus on M81KR use cases

• For bare-metal deployments: the answer is “it depends”

• -Consult the application owner(s)

• -Empirical observation: typically 4 to 6 vNICs; 0 to 2 vHBas

• For ESXi: most of the times 8 vNICs and 2 vHBAs

• -Details in the next slides

• For Hyper-V: most of the times 6 vNICs and 2 vHBAs

• -Details in the next slides

3. Documents

Best Practices Hints and Tips from the field

UCS SAN Deployment Models and Best Practices