Key differences in monolithic vs. microkernel architecture

Microkernel vs. monolithic: The basic difference

The monolithic approach keeps all of VMware's abstracted hardware under one roof. Key parts of the machine, such as drivers, live in the hypervisor, which distinguishes VMware's approach from the microkernel approach.

In Hyper-V's microkernel architecture, Drivers live in a parent partition VM that creates a VM bus framework. Other VMs can then use that bus framework for accessing hardware resources, rather than communicating directly with the hypervisor.

Microkernel and monolithic performance

The monolithic approach offers faster execution for internal messaging. However, a monolithic architecture requires running all resources in the same memory space. Combined with the fact that ESXi doesn't have an additional VM layer that holds the drivers and messaging system, VMware and its monolithic approach should theoretically be faster than Hyper-V. A monolithic architecture has a smaller code base even if the overall package itself is bigger, which also lends itself to speedier performance.

However, you can use a variety of metrics to measure performance stats, and a variety of factors play a role outside of the kernel architecture. Management tools and other parts of the architecture such as networking and storage often have a greater effect on performance than the kernel architecture.

Stability of different architectures

Because of its all-or-nothing style design, monolithic architecture lacks the stability of a microkernel approach. In monolithic designs, kernel panic or other service issues can lead to the entire host crashing. However, a reboot often fixes most issues.

A monolithic design means you might need to replace the entire hypervisor in the event that you add additional features or functions to your system. ESXi has a small footprint, which means reinstallation usually doesn't take a long time, but during updates or upgrades, it can still create additional hassle.

Microkernel architecture separates services from the kernel, which means that part of your service can crash while the system at large remains intact. The microkernel's isolated design also enables you to expand without affecting the kernel, so you can easily add additional services as needed with minimal trouble. However, in a microkernel architecture, failure at or near the hypervisor layer can sometimes result in a required host reboot. You can recover at the hypervisor layer, but recovery often requires additional runtime without a reboot. Still, you can use that additional time to move or properly shut down workloads before a host reboot.

Which architecture is best?

Both VMware and Hyper-V claim their hypervisor architecture leads in performance, security or stability. The simple fact remains, though, that both designs are Type 1 hypervisors. Each design has its current and future benefits and its drawbacks.

VMware's monolithic ESXi hypervisor is mature and well-tested in demanding environments. It has proved its efficacy over and over. You can mitigate its shortcomings by placing the right systems and infrastructure around it.

Although Hyper-V is newer to the marketplace compared to ESXi, its microkernel architecture shows a different way of thinking about the hypervisor's role in a virtual system. Depending on the direction Microsoft takes it, microkernel architecture could become as mature and effective as monolithic hypervisors in the future.