SpringOne Platform 2016
Speaker: Alois Reitbauer; Chief Technology Strategist, Dynatrace
The cloud has transformed the way we build applications. Early adopters prove that the benefits manifest in delivering value faster to the customer, less operational costs and more productive teams. The interesting question is how to get there, especially if you cannot simply start over and get your current applications benched. Through numerous customer engagements we have learned that there is a pattern followed by companies that do this successfully. This talk show you:
-How to deploy faster without breaking things
-Start decoupling a monolith without breaking your business logic.
-Help developers build application using a new paradigm
-Dynamically scale your applications to save costs
-Manage highly dynamic larger-scale micro services without increasing operational costs
As we will walk through the journey we share learning covering challenges and possible solutions on the organizational, development and operational side. As a company that has gone through this transformation itself while onboarding over 500 new customers we have some interesting stories to share.
12. Control code complexity
N+1 interaction patterns - control flow design
Duplicate interactions - bounded context, SPOF
More than 10 service hops - hard to debug
Highly asynchronous - hard to understand
Unusual Log Activity
14. Stage 2: Refactoring towards Micro Services
First micro services
Scheduling and Orchestration
Architectures become polyglot
Independent Release Cycles
16. Control Environment Complexity
Service Call Payload - < 500 kb
Chatty services - > 50 service calls
Inter-tier heavy interactions - granularity and locality
Status Codes - only Ok is 200
20. Watch your application live
Health monitoring – symptoms
Orchestration monitoring - causality
Application monitoring - root cause
21.
22. 7:00 a.m.
Low Load and Service running
on minimum redudancy
12:00 p.m.
Scaled up service during peak load
with failover of problematic node
7:00 p.m.
Scaled down again to lower load
and move to different geo location
Deployments are no longer static
23. You don’t fly by hand here
820 Billion
dependencies
Network Problem
Mushroom cloud effect
It’s also about how you compute within your data center or cloud. As you move toward apps of innovation you will also move toward software-defined data centers. If you’re using the public cloud everything is soft-wired already, but maybe in your data centers you’re looking toward VXLAN or other means to turn your current hard-wired networking into software. And not only networking, but also how you provision systems, how you make them accessible on a self-serve basis to groups like your engineering teams. In many ways your data center is turning into an API and you’re becoming a cloud provider.
Discussion questions:
Are you moving toward self-service for your engineering teams?
Do you see your data center team as becoming a cloud operations team?
B2W example.
9 applications are HEALTHY!!!! only services have issues, but thats fine.
3300 servics running on 10k JVMs
on just 142 hosts
thats no longer for human doing visual monitoring; ist ideal for intelligent monitoring systems to deal with