CHT Project
Progress Report
10/07 Simon
CHT Project
Develop a resource management sche duling algorithm for CHT datacente r.
◦Two types of jobs, interactive/latenc y-sensitive and batch/computation-int ensive.
◦Minimize SLA violation with limited r esources.
# of servers
◦Based on Red Hat Openshift
Current Plan
Two components
◦Scheduler
Deploy container/pod to server
Inside Kubernetes
◦Rule Engine
Decide the number of container/pod for eac h service.
Output the results in JSON
Input of the scheduler
Scheduler
Implement a new scheduler as plug- in and replace the original one.
Use the original scheduler, but ch ange the policy.
◦“Provide a JSON file that specifies the predicates and priority functions to configure the scheduler”.
◦Change the weight of the (built-in) p riority function to meet our score fu nction.
Rule Engine
Decide the number of container/pod for each service according to the monitoring data.
Add/adjust rules to make better de cisions.
However,
◦Rule Engine is another Red Hat produc t.
◦Creating and adjusting new rules requ ires experiences.
Alternative Way
Build our own Resource Allocator.
◦Decides the number of container/pod f or each service according to the moni toring data.
◦Basic rules
Rules about the critical resources such as CPU, memory …etc.
Possible future extension
Apply machine learning to add/adju st the rules.(CHT)
Apply machine learning to minimize the size of a container/pod.(Prof.
Lin)
Next
Study the current predicates and p riority functions inside Kubernete s.
Keep working.
