Lecture 02 - Introduction to Computing Virtualization
Course: Cloud Computing Technologies
Professor: Fulvio Risso, Politecnico di Torino
Source: Introduction to Computing Virtualization, 29 slides
Lecture date: not stated in the PDF.
Numbering: 02 comes from the supplied filename, not a lecture number printed in the slides.
Topics
- Motivation: isolation requirements and underused multicore servers
- Hardware sharing, consolidation, energy, and agility
- COTS hardware and OEM/ODM economics
- VMs, host/guest operating systems, and hypervisors
- x86 virtualization approaches and the transition toward cloud computing
What the professor covered
Why virtualization returned (slides 3-7)
Servers moved from scattered company installations into datacenters, making management easier but exposing widespread underutilization. The historical one application per server rule avoided conflicts that an ordinary shared OS did not sufficiently isolate:
| Isolation requirement | Slide example |
|---|---|
| Configuration / shared components | Applications require different library versions or vendor-certified OS/patch combinations |
| Temporal / performance | One application’s CPU or network usage affects another |
| Spatial / security and reliability | A failure can affect another service; local communication may bypass an external firewall |
Multicore machines exacerbate wasted capacity when applications are single-threaded and cannot be consolidated.
Computing virtualization shares CPU, memory, and I/O among separate OS environments. The lecture describes lift and shift as separating software from its physical hardware and moving it unmodified into virtual servers.
Slide 7 presents a historical timeline: mainframe virtualization around the 1960s, less emphasis with cheaper personal computers, renewed interest in the 1990s, and the lecture’s focus on x86. These are the slide’s broad historical framing.
Related: Computing Virtualization, VM Isolation and Server Consolidation.
Power and cooling (slides 8-12)
A server consumes power even when idle: memory, motherboard, NICs, storage, fans, and the power supply remain active. Slide 8 shows a substantial jump from sleep to idle and a smaller increase as CPU utilization rises. Slide 9 compares power (W) against PassMark score, not CPU utilization; efficiency differs across the plotted machines.
The lecture reports typical per-server CPU utilization of 5-15% and suggests consolidating workloads onto fewer active servers, while leaving enough capacity to avoid resource contention and degraded QoE.
Historical/example figures in slides 11-12 include:
- A 10GbE NIC at approximately 15 W and a 1 TB disk at approximately 10 W idle / 15 W read-write, labeled 2008.
- CPU consumption around 100-300 W and an idle server potentially using 50% of peak power.
- Approximately 10 kW per rack and a 25 ft² footprint, yielding 400 W/ft²; slide 12 labels its density example as 2008 data.
- Facility-power multipliers of about 3 in older datacenters, 1.8 for a conventional datacenter, and 1.2 for a hyperscaler, as presented by the professor.
These figures illustrate the lecture’s reasoning; they are not verified specifications for current equipment. See Server Power Consumption and PUE for the PUE formula and a worked study example.
Benefits, limitations, and usage scenarios (slides 13-18)
Isolation separates service environments; consolidation shares physical hardware among different OSes, improving utilization and potentially reducing operating costs.
Flexibility and agility include pausing/restarting OS execution, migrating VMs even while running, duplicating VMs, disaster recovery, and rapid deployment. The professor considers agility especially important: programmable infrastructure lets autonomic systems react and adapt. AI may decide how to respond, while the infrastructure supplies the ability to change.
Limitations include extra OS resource requirements and difficulty exposing heterogeneous or specialized hardware such as GPUs and offloading cards.
The lecture contrasts server virtualization with workstation/desktop virtualization and treats server virtualization as the stronger economic driver. Slide 18 lists ESXi, Hyper-V, XenServer, and KVM for servers; VirtualBox, KVM, and VMware Workstation for desktops. These are the slide’s examples, rather than a current licensing or product comparison.
Related: Virtualization Agility and VM Migration, VM Isolation and Server Consolidation.
Commodity hardware and the server market (slides 19-23)
Virtual servers can be created with the requested CPU, RAM, disks, and NICs, reducing the need to buy a separate physical machine tailored to every application. Equivalent physical servers can form a resource pool: computing hardware becomes a commodity.
Slide 21 shows worldwide server-company revenue shares from 2018 Q1 to 2020 Q2, with a September 2020 IDC reference. ODM Direct ends at 28.8% in the chart. Slide 23 cites an IDC June 2020 report on the 2019 ODM Direct market: $21.2 billion in sales, with seven vendors accounting for 89%. These are historical market examples.
Slide 22 distinguishes OEM and ODM by who designs the product. Slide 23 discusses white-box purchasing by large datacenter operators.
Related: COTS Hardware and OEM vs ODM.
VM architecture and resource control (slides 24-26)
- VM: a software-defined machine running a guest OS and applications against virtual hardware.
- Guest OS: the OS inside the VM; in the initial transparent-virtualization model it should behave as though on physical hardware.
- Host OS: the OS on the physical machine; the slide’s diagram places the hypervisor within its software layer.
- Hypervisor / VMM: the software virtualizing CPU, memory, and devices.
The hypervisor partitions resources where possible, such as disjoint memory regions, and arbitrates shared resources where necessary, such as a single physical NIC.
Slide 26 describes a frequently stripped-down, Linux-based host environment, native hardware drivers, and standard virtual devices for guests. The host remains an attack surface, despite potentially being smaller than a general-purpose OS.
Study clarification: this is the architecture introduced by these slides, not a claim that every hypervisor requires a conventional host OS. The transparent guest description also needs to be read alongside paravirtualization on slide 27.
Related: Virtual Machines and Hypervisors.
x86 techniques and what comes next (slides 27-29)
| Slide terminology | Slide date / example | Main distinction |
|---|---|---|
| Dynamic Binary Translation | 1999 / VMware | Guest OS runs unmodified |
| Paravirtualization | 2003 / Xen | Guest OS must be modified |
| Hybrid Virtualization | 2006 / all | Exploits host OS primitives and CPU/chipset support |
Keep the professor’s term hybrid virtualization for this comparison. These dates are the slide’s timeline, not an exhaustive history of each technique.
Virtualization also applies to embedded environments with isolation and compatibility needs; slide 28 gives consolidation of car controllers as a possible use case.
The conclusion points toward lighter virtualization using containers and the need for cooperation among multiple physical servers, leading into cloud computing. Containers and distributed cloud management are signposts here, not fully developed topics.
Related: x86 Virtualization Techniques, Computing Virtualization.
Questions to check understanding
- Why is installing several applications on one shared OS different from using separate VMs?
- Why can consolidation reduce power even if the remaining servers become busier?
- What does PUE measure, and what does it leave out?
- Why is agility more than a hardware-cost saving?
- Which resources can be partitioned, and which need arbitration?
- Why does paravirtualization qualify the claim that guest software is unmodified?
- What extra problem appears when virtual servers must cooperate across physical hosts?
Source
Next: Lecture 03 - Computing Virtualization Technologies and Tools develops CPU, memory, and I/O virtualization, Linux tools, migration, and nesting.
Fulvio Risso, Introduction to Computing Virtualization, Politecnico di Torino, slides 1-29. PDF page numbers match slide numbers. Original: 10 Courses/Cloud Computing Technologies/Resources/02 - Intro Computing Virtualization.pdf. Sections explicitly marked as study clarification add explanation to the supplied material.