* Things to consider when designing a data center network
“Put all your eggs in one basket – then watch that basket.” The quote is usually attributed to Mark Twain, who was making the point that for some things, distributing isn’t always the best strategy. Sometimes consolidation is a better approach, a lesson many data center managers have internalized over the past few years.
Most of the companies with which Nemertes has spoken have drastically reduced the number of data centers they support. Typically, most organizations have reduced the number of data centers from double digits down to an average of two to five data centers per organization.
Consolidation is great from the perspective of reducing overall management costs, improving standardization and enabling the deployment of technologies – such as virtualization and storage-area networking and network-attached storage – that cut costs while improving availability. But there are some non-obvious consequences of data center consolidation, such as increased vulnerability to disasters, facilities challenges and overall greater capacity requirements.
In previous newsletters, we’ve discussed these issues. But one critical topic that’s worth bringing to the fore is the networking requirements for the consolidated data center.
In designing data center networks, there are two key areas to consider: the LAN and the WAN.
Obviously, data center LANs require greater capacity than ordinary office LANs. Enterprises may still be debating the merits of Gigabit vs. 100M bit/sec Ethernet in office environments, but for data centers, Gigabit and 10 Gigabit Ethernet designs are a no-brainer.
A less-obvious LAN requirement is quality of service. At the same time that companies have been consolidating their data centers, they’ve also been consolidating applications (sometimes from 1,000 or more down to a handful). That means multiple mission-critical applications may be residing on data center servers and sharing a common network infrastructure. Ensuring that each application gets the quality-of-service it needs requires an end-to-end design philosophy.
High-quality security is a must. We’ve written reams about the need for virtualized security (such as that found in the products provided by Inkra, Cisco and others), which enables defense in depth and helps contain any potential breaches and limit their impact on organizations.
It’s also important to remember facilities security. It’s not overkill to deploy biometrics (such as retinal scans, face-recognition technology and fingerprint scans) to control and manage access – not when a breach could conceivably take down a company.
As for the WAN, there are three primary issues to consider. First is bandwidth: Most data centers require a minimum of T-3 connectivity, and the majority of folks we work with put data centers on OC-3 and up.
The second issue is redundancy. Multi-homing is critical (and it’s worth considering approaches such as InterNap’s that make it painless). So is physical circuit redundancy.
Finally, data center managers need to think about how to enable remote sites to transition from one data center to another in the event of an outage. This is harder than it looks because the IP VPNs that organizations are increasingly using to connect some or all of their remote sites often require reconfiguration.
The bottom line? Data center networks are different – and it’s worth keeping a close eye on that particular basket.
