Zero Downtime Migrations TECHNICAL GUIDE
By Jonathan Baier & Prakash Patil
Working with Cloud Technology Partners application migration practice we see a lot of different enterprise application portfolios and have uncovered some consistent patterns . Most organizations moving to cloud have various levels of legacy applications , with some lending themselves well to cloud environments , while others require significant refactoring . When applications are mission critical , a business cannot afford long periods of downtime during the migration . In these cases , we often implement a zero downtime migration .
While zero downtime migrations are great for relieving business interruptions , they should still be approached with caution . Additional setup work , as well as tightly coordinated cutovers , add to the total time and cost compared with traditional “ lift-andshift ” migrations .
However , the bigger issue is that a request for zero downtime can often be an attempt to shortcut the migration and discovery process entirely . The thought here is that if replacement servers can be built and initially tested , teams will not need to spend additional time pouring through old systems . It turns out that the opposite approach is true .
For those applications that have a high enough business value and impact to merit zero downtime , even more time needs to be spent on the discovery and dependency mapping . Even if duplicate systems are stood up and tested in parallel to the existing system , it ’ s often hard to get complete test coverage before cutover . As they say , “ There is nothing like running in Production except running in Production !”
Of course , there certainly are cases where the application ’ s impact is very high and the additional work is therefore well worth the investment .
The rest of this article takes a deeper look into the challenges we faced with a recent zero downtime migration .
Migration Background
As featured in last quarter ’ s Doppler magazine , AVID Technologies has migrated the bulk of its infrastructure from a fully managed private datacenter to AWS . Most of the applications were COTS ( Commercial Off The Shelf ) and good candidates for a lift-and-shift migration .
However , there were two mission critical applications that required migration with no downtime , as outages would have had significant business impact . Both applications also had a high number of dependencies from other applications and systems .
The overall migration of these two applications addressed many challenges including :
• Gaps in configuration management knowledge
• Provisioning HAProxy based load balancers instead of using AWS ELBs ( Elastic Load Balancers ), due to compatibility issues
• Syncing almost 2TB of data and keeping it in sync for the duration of the migration .
8 | THE DOPPLER | SPRING 2016