Hypercallbacks: Decoupling Policy Decisions and Execution

摘要

While hardware assisted virtualization provides relatively high security, its performance lags behind alternative approaches, such as containers. Existing paravirtualization techniques do not suffice as they require context switches that induce high overheads. We have shown hypercallbacks, which can run efficiently in the hypervisor, may reduce virtualization overheads, and have demonstrated a prototype using memory management. We have also shown that the requirements for hypercallbacks differ from previous work on safe OS extensions, and denning those requirements will be necessary to extract the maximum performance from hypercallbacks.