Saturday, 10 August 2013

Why Active / Active DC Design is not preferred by many yet?


We all know that Active / Active is always there as an option for a Data Center design. But We could not find many actual implementations out there. While everyone think that the Active / Active DC solution maximizes the DR investments, not many are getting in there. Can you highlight the challenges and issues that holds this option back?

By: David Matt


There are three key issues or challenges that when addressed will result in higher investment than expected and / or increase the complexity. The three challenges are explained further as below:

  1. Network Latency - To be Active / Active, the data created in on DC should also be created in the other DC with least latency, so that the user requests land on either DC would happen to see the most current data. Though there are tools and technologies to offer active / active solution, they are constrained by a minimum network latency between the DCs. For instance, EMC offers VPLEX based active / active DC solution, but requires that the latency be 50 ms or less. Considering the speed of light and the availability of the optic fiber channels, theoretically, this is achievable within reasonable distance, but in practice, there are various other factor that would have impact on achieving the required latency consistently.
  2. Network capacity - When we talk about active / active data centers, it is not just the database replication, it also requires the application state and associated data to be replicated in real time. More over, given the actual fluctuates, there is a need to provision for the peak time capacity so as to ensure that the latency does not go down in peak periods. Thus, the network capacity needs can easily be three times or even four times of that needed for an active /passive DC design.This need will straight away beat the assumed cost advantage of using the DR site for production as well.
  3. Overall DR Capacity - Assuming that the DR site will also be active for read and write, there is a tendency size the Primary and DR sites to half or little over half of the planned production capacity so that both put together would meet the production demands. Now, in such a scenario, if the either of the site fails, the production capacity comes down to half leading to ineffective DR.

 Given the above key challenges and other related issues, organizations end up with Active / Passive DC solutions or a hybrid solution, where specific applications are partitioned to run out of specific Data Center.