Data Center Architecture and Design Principles
Domain 1 - Data Center Architecture and Design Principles
This domain tests whether you can turn business and workload requirements into a defensible physical and logical data center design. Expect questions on leaf-spine topology, oversubscription, redundancy, failure domains, and capacity planning rather than product SKU memorization.
Core rules
- Design for east-west traffic: most modern DC flows are server-to-server, so a leaf-spine fabric with predictable hop count beats a traditional three-tier design.
- State oversubscription explicitly: access bandwidth divided by uplink bandwidth. A 2:1 ratio is tighter than 4:1. Know both the intended ratio and the ratio after one link or one spine fails.
- Separate failure domains: dual power feeds, redundant supervisors, diverse uplinks, and management out-of-band. A design that survives one link but not one spine is only partially redundant.
- Size for growth: reserve leaf ports and spine capacity for a defined growth period; do not consume every uplink port on day one.
Worked example
A rack has 48 servers at 10G each, so access bandwidth is 480G. The requirement is no worse than 3:1 oversubscription. Uplink capacity must be at least 480G / 3 = 160G. Using 100G leaf uplinks, two uplinks give 200G, or 2.4:1 - acceptable. If each leaf connects to two spines, one 100G link per spine, losing one spine leaves 100G, or 4.8:1. If the business requires 3:1 even after a spine failure, you need three 100G uplinks per leaf across three spines, or 300G normal and 200G after one failure (2.4:1). Show this arithmetic; it is the mark-earning step.
Traps
- Counting ports instead of bandwidth.
- Quoting only normal-state oversubscription and ignoring failure.
- Forgetting out-of-band management and console access.
- Mixing power redundancy domains with network redundancy domains.