- Core Isolation - Enforce maximum space usage limits
- Limit I/O bandwidth usage- Ports
Resource allocation
How to prevent frameworks starving?
Before workload After workload
How to prevent frameworks starving?
Static reservation - Good for Stateful services
Dynamic reservation - Good for Stateful Scheduled tasks
Resource preemption
Resource preemption
Frameworks: Marathon!
Built to support long running jobs
Specificing contranstraints
Good to ensure tasks are:
● Running all slaves● Running At least one per
datacenter● Running on specific slaves● Running at least one per
cluster group
Scaling applications out
New tasks can be allocated dinamically
Handling failures
Handling failures
Timeout!
Handling failures
Report the frameworks!
Handling failures
Reschedule the tasks!
● Automation is heavily needed on this environment● Troubleshooting can be tricky● Monolithic systems may not take full advantage from this solution● Ops: supporting the platform rather than specific products● Applications that demands specific SO and hardware improvements may
wont leverage the benefits from this approach
Considerations
Managing Computational resources with Apache Mesos