This is a really fascinating problem set for distributed computing. I've been involved in Active-Active designs that never fully went live, and some Active-Passive, and I def agree with the criticisms of Active-Passive, and the assertion that you should fail back and forth regularly to ensure that you have a procedure that works, and resources that actually match.
It's easy for N+1 to be N+0.8, and at peak load, that may be essentially worthless.
Exactly justizin! I find it is actually very difficult to ensure we have not crossed over that threshold of N+X where X < 1. You have to test each application tier after each release to be sure of this, and for app tiers that have data storage, the results are only relevant when nodes are loaded to their maximum target data load and in their worst case "availability model degraded" state. (I am the author of this blog post BTW)
2 comments
[ 2.7 ms ] story [ 18.5 ms ] threadIt's easy for N+1 to be N+0.8, and at peak load, that may be essentially worthless.