CS 695: Virtualization and Cloud
Computing
Lecture 1: Introduction
Mythili Vutukuru
IIT Bombay
Spring 2021
Virtualization and cloud computing
• What is the cloud?
• Commodity servers with lots of compute and storage, connected with high
speed networking, located in data centers
• What is virtualization?
• Multiple virtual machines (VMs) can run inside a physical machine (PM)
• VM gives user an illusion of running on a physical machine
• Containers are like lightweight VMs
• Virtualization is a building block for cloud computing
• Virtualization enables multiple clients share the cloud’s compute resources
• Multiple users on VMs/containers can share same cloud server
• In addition to compute, clouds also manage large amounts of data
• Cloud storage/big data systems for efficient storage and retrieval of data
What is this course about?
• Two parts of the course:
• Understanding how virtualization works
• Basics of cloud storage frameworks
VM VM VM VM
Cloud compute
VM VM VM VM (servers in a
data center)
Cloud storage over
memory/disk
Why cloud computing?
• Public cloud providers (Amazon AWS, Microsoft Azure, Google Cloud
etc) setup and maintain data centers with high-end servers
• Powerful CPUs, lots of memory, disk storage etc., available to users
• Organizations can also run a private cloud only for their users
• Why run applications on cloud and not on “bare metal” servers?
• Multiplexing gains: multiple VMs can share the system resources
• Lower overhead of maintenance: hardware/software maintained by providers
• Flexibility: VMs can move to another machine if one fails
• Pay as per usage: no need to invest in servers if only lightly used
• Disadvantages of running applications on cloud
• Performance: longer delay to access servers via internet
• Higher cost if heavily used
Virtualization terminology (1)
• We will study system virtualization, or how to run one full system (OS
and applications) over another OS
• We do not cover process virtualization (e.g., Java virtual machine) which lets a
single process run on a different architecture from underlying machine
• Hypervisor or virtual machine monitor (VMM): a piece of software
that allows multiple VMs to run on a physical machine (PM)
• We will study how VMMs are designed
Virtual Machine
Hypervisor/VMM
Physical machine
Virtualization terminology (2)
• Guest OS runs inside the VM, and host OS runs on the PM
• Type 1 hypervisor: runs directly on hardware, no need for host OS
Virtual Machine
Type 1 Hypervisor
Hardware (CPU/RAM)
• Type 2 (hosted) hypervisor: runs as an application on top of host OS
Virtual Machine
Type 2 Hypervisor
Host OS
Challenges to virtualization
• Guest OS expects complete control over hardware, but VMM must
multiplex multiple guests on the same hardware
• Understand how operating systems work (prerequisite)
• How to trick the guest OS into relinquishing hardware control?
• We will study the following ways to design virtual machine monitors
• Hardware assisted virtualization (e.g., KVM/QEMU): modern CPUs have
support for virtualization and VMMs are built over this support
• Full virtualization (e.g., VMWare): Original technique to run unmodified OS
over original hardware with no virtualization support
• Paravirtualization (e.g., Xen): Modify OS source code to be compatible with
virtualization
• Understand how CPU, memory, I/O devices are virtualized with each
of the above techniques
Other topics in virtualization
• VM live migration and related ideas
• VMs can moved from one physical machine to another
• Why? Maintenance of machines in the cloud, fault tolerance etc.
• How are VMs migrated without impacting the application in it?
• Use similar techniques for other uses like VM checkpointing
• Containers: lightweight virtualization technique
• Underlying Linux concepts of namespaces, Cgroups
• Container frameworks like LXC, Docker, Kubernetes
Introduction to Cloud Computing
• Architecture of cloud applications: compute and storage options
• Compute in VMs, containers etc.
• Traditional storage in databases, now moving to simpler key-value stores etc.
• Cloud storage techniques
• In-memory key-value stores (Amazon Dynamo)
• Semi-structured data storage (Google Bigtable)
• Application specific storage (Facebook’s Haystack to store photos)
• Caching-based optimizations to the cloud storage layer (Facebook’s memcache)
• We will only briefly touch upon distributed systems concepts like
replication, fault tolerance etc. as required
• This is not a distributed systems course
Summary
• Introduction to virtualization and cloud computing
• Basic terminology and concepts
• Course outline
• Techniques to design VMMs (hardware-assisted, full, para virtualization)
• CPU, memory, I/O virtualization
• VM live migration and related ideas
• Containers
• Cloud applications and cloud storage
