Data Center Architecture
Layered Data Center Architecture
If we subject a network topology design to the universality theorem (for a given number of switches, the most optimal network exists), we quickly conclude that existing network designs, which are optimized for transporting data into and out of a data center, are not the most optimal.
There are numerous data center models. The most widely used is the multi-tier. Multi-tier server farms built with processes running on separate machines can provide improved resiliency and security. Resiliency is improved because a server can be taken out of service while the same function is still provided by another server belonging to the same application tier. Security is improved because an attacker can compromise a web server without gaining access to the application or database servers. Web and application servers can coexist on a common physical server; the database typically remains separate.
We Can Help
Network Architects plan and design computer networks. Define Policy and Procedure, design test plans and publish deployment and provisioning guidelines.
Layered Data Center Architecture
The physical network most often is Layered. Pairs of switches exist in the Core, Distribution and Edge, paired with LACP link bundles, resulting in high resiliency. The downside is that links in core and distribution are generally 40G/100G, increasing costs and power requirements. In applications where speed is important, such as High Frequency Trading, this is not the fastest design as every element a packet traverses adds inherent latency.
Spine-Leaf Data Center Architecture
There is a growing utilization of a Spine-Leaf where speed is more important than network resiliency.
The key characteristics of the Spine-Leaf topology are:
- Multiple design and deployment options enabled using variable length spine with ECMP to leverage multiple available paths between leaf and spine.
- Future proofing for higher performance folded Clos network through the selection of high radix (or port count) switches at the spine.
- Reduced network congestion using a large, shared buffer-switching platform for leaf node. Network congestion results when multiple packets inside a switch request the same outbound port because of application architecture.
- Lower power consumption using multiple 10Gbps between spine and leaf (in a folded Clos topology) instead of a single 40Gbps link. The current power consumption of a 40Gbps optics is more than 10X a single 10Gbps
Spine-Leaf Data Center