of workload mobility needs to be extended large-scale; note that this doesn’t necessarily imply layer-2
extensions across the globe. It simply implies that the workload needs to be moved or distributed differently,
which can be solved with geographically-based anycast solutions, for example.
As discussed in the automation/orchestration section above, orchestrating workloads is a major goal of
cloud computing. The individual tasks that are executed in sequence (and conditionally) by the orchestration
engine could be distributed throughout the cloud. The task itself (and the code for it) is likely centralized in
a code repository, which helps promote the “infrastructure as code” concept. The task/script code can be
modified, ultimately changing the infrastructure without logging into individual devices. This has CM benefits
for the managed device, since the device’s configuration does not need to be under CM at all anymore.
1.6 Compute virtualization
Conceptually, containers and virtual machines are similar in that they are a way to virtualize services/machines
on a single platform, effectively achieving multi-tenancy. The subsections of this section will focus on their
differences and use cases, rather than discuss them at the top-level section.
A brief discussion on two new design paradigms popular within any data center is warranted. Hyper-
convergence and disaggregation are polar opposites but are both highly effective in solving specific
business problems.
Hyper-convergence attempts to address issues with data center management and resource provisioning.
For example, the traditional DC architecture will consist of four main components: network, storage, com-
pute, and services (firewalls, load balancers, etc.). These decoupled items could be combined into a single
and unified management infrastructure. The virtualization and management layers are integrated into a
single appliance, and these appliances can be bolted together to scale-out linearly. Cisco sometimes refers
to this as the Lego block model. This reduces the capital investments a business must make over time
since the architecture need not change as the business grows. Hyper-converged systems, by virtue of their
integrated management solution, simplify life cycle management of DC assets as the “single pane of glass”
concept can be used to manage all components. Cisco’s Hyperflex (also called Flexpod) is an example of
a hyper-converged solution.
Disaggregation is the opposite of hyper-convergence in that rather than combining functions (storage, net-
work, and compute) into a single entity, it breaks them apart even further. A network appliance, such as a
router or switch, can be decoupled from its network operating system (NOS). A white box or brite box switch
can be purchased at low cost with some other NOS installed, such as Cumulus Linux. Cumulus generally
does not sell hardware, only a NOS, much like VMware. Server/computer disaggregation has been around
for decades since the introduction of the personal computer (PC) whereby the common Microsoft Windows
operating system was installed on machines from a variety of manufacturers. Disaggregation in the network
realm has been adopted more slowly but has merit for the same reasons.
1.6.1 Virtual Machines
Virtual machine systems rely on a hypervisor, which is a software shim that sits between the VMs them-
selves and the underlying hardware. The hardware chipset would need to support this virtualization, which
is a technique to present hardware to VMs through the hypervisor. Each VM has its own OS which is
independent from the hypervisor. Hypervisors come in two flavors:
1. Type 1: Runs on bare metal and is effectively an OS by itself. VMware ESXi and Linux Kernel-based
Virtual Machine (KVM) and are examples.
2. Type 2: Requires an underlying OS and provides virtualization services on top through a hardware
abstraction layer (HAL). VMware Workstation and VirtualBox are examples.
VMs are considered quite heavyweight with respect to the overhead needed to run them. This can reduce
the efficiency of a hardware platform as the VM count grows. It is especially inefficient when all of the VMs
Copyright 2020 Nicholas Russo http://njrusmc.net 19