Unit 2: Virtualization and capacity planning

Virtualization creates a virtual version of a resource, such as a server, storage device or network, so that several virtual machines (VMs) can share one physical machine, each with its own operating system and applications. The software that does this is the hypervisor or virtual machine monitor.

• Sharing: many guests use the same host resources.
• Aggregation: many hosts are pooled to look like one large resource.
• Emulation: a guest runs on hardware different from the real one.
• Isolation: each guest sees only its own virtual resources.

At the system level there are three ways to virtualize hardware:

• Full virtualization: the guest OS runs unmodified and does not know it is virtual.
• Paravirtualization: the guest OS is modified to talk to the hypervisor directly, which is faster.
• Hardware-assisted virtualization: processor features such as Intel VT-x and AMD-V help the hypervisor.

Virtualization is the foundation of IaaS. The provider runs thousands of hosts, and each customer VM is created from an image in minutes. It enables resource pooling, elasticity (start and stop VMs as needed), multi-tenancy (many customers on one host in isolation) and live migration (moving a running VM to another host). Pros: better hardware use, lower cost, quick provisioning, isolation, easy backup and snapshots, and simple disaster recovery. Cons: the hypervisor adds a small performance overhead, one host failure affects many VMs, resource contention between VMs, licensing and management complexity, and a hypervisor attack is serious.

• Xen: an open-source Type 1 hypervisor. A privileged control domain (Dom0) manages the unprivileged guests (DomU). It pioneered paravirtualization and underpins early Amazon EC2.
• VMware: a family of products, with ESXi as the Type 1 hypervisor, vSphere for management, vMotion for live migration and Workstation for desktops.
• Microsoft Hyper-V: a Type 1 hypervisor built into Windows Server, supporting live migration and used by Azure.

Capacity planning is estimating the resources an application will need so that performance stays acceptable without paying for unused capacity.

• Elasticity vs scalability: scalability is the ability of a system to handle growth by adding resources; elasticity is doing so automatically and quickly, in both directions, as demand rises and falls.
• Defining a baseline: measure the system under normal load to record what is typical. Without a baseline you cannot tell whether a number is good or bad.
• System metrics: CPU utilisation, memory use, disk input and output, network throughput, and response time.
• Load testing: simulate many users to find how the system behaves as load grows.
• Resource ceilings: the maximum a resource can deliver, such as the top bandwidth of a network link or the CPU limit of an instance. When one ceiling is hit, that resource is the bottleneck.
• Server and instance types: providers offer general-purpose, compute-optimised, memory-optimised and storage-optimised instances; choose by workload.
• Network capacity: check bandwidth and latency between users, servers and storage, because a fast server behind a slow link is still slow.