RightScale Blog

Cloud Management Blog
RightScale 2014 State of the Cloud Report
Cloud Management Blog

Add new comment

Hey Patrick, Nice post! You've raised some great issues, and the multi-tier scheme you've illustrated in figure 8 is a solid design template. Just two quick observations: 1) Regarding TCP load balancer failure, while it's true that most browsers do have reliable "time out and move on" behavior, that can still result in a suboptimal client experience. An alternative to consider might be a DNS load balancer that does active healthchecking of the TCP lb's in the first place, to avoid clients having to time out at all. 2) With a DNS load balancer in place that does that healthchecking, it's possible to deliver single "A" records to clients, instead of multiples. And it's then possible to maintain a "sticky" association between the client and its TCP load balancer instance. For delivering apps that are more stateful, this is an invaluable tweak. Cheers, Jon Braunhut Chief Scientist KEMP Technologies