Skip to content

Latest commit

 

History

History
1 lines (1 loc) · 773 Bytes

File metadata and controls

1 lines (1 loc) · 773 Bytes

Resource under-utilization is common in cloud data centers. To improve server utilization, batch workloads can be run in the background to leverage idle resources. However, customer Virtual Machines' (VMs) performance can be degraded because of resource contention. This is more challenging since customer VMs are black-box and their resource usage is dynamic and unpredictable. We have been working on Scavenger, a batch workload manager that opportunistically runs containerized batch jobs next to the customer VMs to improve utilization without impacting the VMs' performance. Please see ipc_monitoring_cpu_allocation_v6.cpp that implements our reactive algorithm to monitor IPC of a foreground VM and chnage the allocated CPU_quota to the batch container accordingly.