A virtual machine (VM) is a fully-fledged, standalone operating environment running on a physical computer. Unlike the host computer its running on, a VM is not a physical machine, thus the designation of virtual. It does, however, have all the components of a physical computer system (CPU, RAM, disk, networking, and operating system) encapsulated in one or more files. As a result, it can process data just like any computer and, for all intents and purposes, be accessed by users while the physical machine provides the underlying hardware resources.
However, VMs cant directly access these hardware resources. Theres a layer of abstraction between a VM and its physical host. This abstraction layer is called a hypervisor. A hypervisor is a specialized software program that runs on the physical host and interacts with both the host machine and the VM, abstracting the host computers resources to the VM. The VM, on the other hand, thinks the resources presented to it by the hypervisor are coming from a physical machine.
Virtual Machine General Architecture (type-1 virtualization)
Its typical to run multiple VMs on the same machine, so VMs can share the CPU, RAM, disk capacity, and the network bandwidth of the host computer. However, each VM only operates with the resources presented to it, completely unaware if other computers are accessing the resources its sharing. This isolation ensures that a VM cant change the underlying settings of the host and, therefore, cant affect other VMs.
Since VMs are actual computer files, they can be snapshotted, replicated, copied to other physical machines, or even deleted.
In this article, we will discuss VMs, the different types of virtualizations, where VMs are most useful, and where they are probably not the best option. We will see their evolution from the on-premise data centers to the cloud, and then to the next level of abstractioncontainers. VMs are still used for running service-based applications similar to containers, and we will finish this article with a discussion about how VMs can be made part of a service mesh.
Depending on the type of hypervisor, there can be two types of virtualizations:
Full virtualization, or type-1 virtualization, involves running a hypervisor on a bare-metal server (such as directly on top of the hardware level as standalone software). In this case, the hypervisor acts as a specialized operating system for the physical host. This is the case with most corporate virtualization scenarios. Some examples of such hypervisors include VMWare ESXi, Citrix XenServer, Redhat KVM, or Microsoft Hyper-V. The above image represents type-1 virtualization.
Type-2 virtualization involves running a hypervisor on top of an already running operating system (OS). In this case, there are two levels of abstraction for the VMone from the hypervisor, and the other from the OS of the physical computer. The OS abstracts the hardware resources to the hypervisor, and the hypervisor then abstracts that to the VM. A typical example of type-2 virtualization is VirtualBox, or VMWare Workstation Player running on a Windows laptop.
Virtual Machine Use Cases
Enterprises started to widely adopt virtualization technologies in the mid-2000s as companies like VMWare helped demonstrate the simplicity and cost savings they offered. This ultimately paved the way to an even more revolutionary change: the cloud. Today, some of the main reasons organizations use VMs include:
Running different types of workloads in the same physical server in an isolated way.
Implementing optimal resource utilization of physical servers, thereby saving space and hardware cost.
Allowing engineers to have their own development environment in their local workstations.
Creating sandboxes for security, for testing unknown or new applications, or for analyzing malware.
Developing and testing Standard Operating Environment (SOE) for both workstations and servers.
Creating snapshots for disaster recovery and backup purposes.
Virtual Machine Anti-Patterns
There are certain use cases for which VMs may not be the best solution, including:
Managing large, monolithic, or legacy applications running on-premise that don't support virtualized environments.
Using older versions of software that dont support virtualized environments.
Using applications that require specialized hardware dongles or Hardware Security Modules (HSM) to physically attach to a machine to operate.
Using software thats not licensed for use on virtual machines.
Running clustered applications that require precise synchronization across all nodes, as VM clocks often dont synchronize well with the host clocks.
Hosting data that is too sensitive for VMs running on a shared tenancy. These applications run best on dedicated servers on-premise.
Virtual Machines in the Cloud
Virtual machines are the main computing resource in any cloud environment. Customers can spin up, spin down, scale up, or scale down any number of these VMs to run their workload in the cloud. AWS, Azure, GCP, and DigitalOcean all have their version of a VM service. But in all cases, these VMs all run on physical machines hosted in the cloud providers data center. The user only pays for the VMs for the duration of the usage.
Cloud VMs are an example of Infrastructure-as-a-Service (IaaS) and are the more recognizable form of virtualization. Other cloud models like Platform-as-a-Service (PaaS), Software-as-a-Service (SaaS), and some newer models like Database-as-a-Service (DBaaS) and Function-as-a-Service (FaaS) all make use of VMs. For example:
Managed services like Amazon RDS, Redshift, EMR, Azure Synapse, or Google BigQuery use virtualization technologies to create database instances and achieve massively parallel processing (MPP) capability.
Serverless functions running on AWS Lambda, Azure Functions, or Google Cloud Functions all run on a server somewhere behind the scenes. And that server is virtualized.
SaaS solutions like Office 365, Google G-Suite, or Snowflake data warehouse use VMs to provide a consistent customer experiencethe user just doesn't get to see the VMs that power these platforms.
Single and Multi-Tenant Cloud VMs
Cloud VMs run on physical machines, but these machines often host multiple customers VMs. When spinning up a VM, theres no way to know which physical server it will be created on, nor what other VMs share that physical machine. This is the default behavior known as multi-tenancy.
A Multi-tenant Setup
Despite the ample controls to ensure VMs cant see each other, this can be an issue for security or performance-conscious customers. Perhaps for data sensitivity reasons or network concerns, an organization might want physical servers dedicated to its VMs alone. When a cloud customer spins up their VMs in a dedicated host, its called a single-tenant host.
A Single-tenant Setup
Cloud vendors like AWS provide options to select either single or multi-tenant setup when creating a VM:
AWS EC2 Tenancy Options
Opting for single tenancy VMs costs more. However, if your workload needs high security and isolation, its the better option.
VMs versus Containers
Containers are a layer of abstraction above both VMs and physical machines. While a VM abstracts a complete machine, a container only abstracts an application and its dependencies.
A container is a packaged unit of a running application and all its dependencies, irrespective of where its running.
Containerized Application Architecture
By itself, a containerized application starts as an image filea standalone, lightweight text file with instructions on what other dependencies must be installed and configured to create a fully running application. As a text file, its easy to port across different environments. A container is spun up by a container runtime engine. As long as there is a container runtime available for the machines (physical or virtual) OS, the containerized application will run exactly the same everywhere.
Containers became popular for running microservices. In large distributed applications, microservices perform small, atomic operations. By running individual services within containers, applications become more portable, fault-tolerant, efficient, and performant.
By nature, containers are ephemeralnot saving data locally; rather, they save it to external storage media mapped to it as standard volumes. So, even if the container crashes, no data is lost. When a container orchestrator like Kubernetes is used, this becomes even more efficient. With Kubernetes, the entire process of managing network connectivity, load balancing, and scheduling of containers becomes automated.
Kubernetes runs containers within one or more of its pods. A pod is the smallest logical unit of computing resource. If a running pod crashes or becomes unresponsive for some reason, Kubernetes simply starts another pod with the container within it, and the container application picks up where it left off.
Generic Kubernetes Architecture
Containers present some unique benefits over VMs:
Compared to VM image files, which are bulky and not easily portable across different OS or hypervisors, container image files are easier to port.
With a container orchestrator, changing image versions for a container doesnt require network reconfiguration and has very little downtime. For a VM, changing the machine image, and getting new machines to use that image can be an involved, time-consuming process.
In cloud environments, you can manually create clusters to run containers or let a managed service like AWS Elastic Kubernetes Service (EKS) create it for you. Theres no such service for creating a cluster of VMs.
Networking Virtual Machines to Run Microservices
Just like physical computers, VMs also need to be part of computer networks to perform most workloads. For example, a physical server may host four virtual nodes of a cluster, each needing to communicate with the other. Ensuring connectivity between all four VMs will allow the application to work cohesively.
In fact, VMs can represent an entire functioning networkmade possible either through the networking feature of the underlying hypervisor, or having the VM connect through the host's network adapter to be part of a VLAN. This is the case with cloud VMs, which are created within Virtual Private Clouds (VPC). VPCs are virtual networks with their own IP CIDR range, subnets, and network ACLs. They also can implement firewall rules with Security Groups.
Using the VPC networking facility, a cloud VM can be configured to stay private, so it can't access the internet, but can communicate with other VMs within the cloud account. Or, a cloud VM can be public, where it can send traffic to and receive traffic from the internet.
It's also possible to achieve connectivity between VMs across network boundaries. Many organizations have both an on-premise, private cloud in their data center and a public cloud footprint. This is called a hybrid setup. Typically in a hybrid setup, the on-premise and the cloud network is connected through a VPN or a dedicated, high-speed direct link like AWS Direct Connect.
Sometimes organizations can also use multiple cloud providers to run their workloads. This is known as a multi-cloud setup. For example, an enterprise may use Azure AD for centralized authentication, and configure services running on their AWS account to authenticate through it. Connectivity in a multi-cloud setup is usually achieved through the public internet.
Problems when Running Microservice Applications in Hybrid and Multi-Cloud
Hybrid cloud setups allow enterprises to keep their on-premise virtual machines, while at the same time taking advantage of the public cloud's elasticity. With a purely multi-cloud setup, organizations can save capital and operational expenditure of a private cloud while making use of the public cloud's elasticity. However, both these approaches come with their own challengesparticularly when service-oriented applications span across networked environments.
Despite the rise of containers and container orchestrators, many organizations still run their microservices on virtual machines. When these distributed services try to communicate with one another, they face the risks associated with slow, unpredictable, or insecure networks.
Even when services could be loosely coupled with message queues, there still needs to be a reliable way for them to communicate. Creating highly available, intelligent routing mechanisms for service communication can be a cumbersome task. Similarly, ensuring a consistent network ACL for these VMs across public and private clouds can be difficult.
Kuma is an open-source CNCF service mesh, originally developed by Kong. Organizations can use it in any microservice implementation, including cloud-native Kubernetes cluster, legacy on-premise virtual machines, or even a hybrid mixture of the two. This is the reason its known as a multi-tenant, universal service mesh.
Kuma frees up individual microservices from having their own service-to-service communication logic. This connectivity and routing logic is abstracted away in a series of software-based network proxies. Rather than microservices directly communicating with one another, their requests are handled by sidecar proxies which then take care of the routing.
The Data Plane of Kuma Service Mesh in a Hybrid Cloud
As you can see from the image above, Kuma can be seamlessly deployed in a multi-zone setup. A zone is a deployment environment like a Kubernetes cluster, an on-premises cluster of virtual machines, a VPC, its subnet, and so on. Kuma can easily synchronize the service connectivity and service discovery across different zones.
As Kuma centralizes this connectivity component, any changes to the connectivity rules can be applied from a single plane: the control plane. The control plane receives user input, creates and configures service meshes, adds services to the mesh, configures their behavior, and applies policies.
Kuma is built on top of Envoy and runs a sidecar alongside every service instance, as shown in the image above. Its job is to process incoming and outgoing service requests.
Typical administrative tasks like updating mutual TLS across all tenancies can be easily done with a single policy update.
As we have seen, VMs are widely used for several use casesfrom developer workstations to cloud-hosted clustersand for their simplicity and cost-effectiveness. Although container orchestration has taken application high availability to a new level, theres still room for improvement when distributed applications run in hybrid or multi-cloud networks and use VMs.
Using a proven and advanced service mesh control plane like Kuma can guarantee connectivity, security, scalability, performance, and seamless integration for these cross-boundary applications.