VM for cloud infrastructure

1. 6-ON THE MANAGEMENTOF
VIRTUAL MACHINES FOR
CLOUD INFRASTRUCTURES
Cloud Computing
Principles and Paradigms
Cloud Computing - Part II 1
27th Jan, 20201

2. In 2006, Amazon started offering virtual machines (VMs) to
anyone with a credit card for just $0.10/hour through its Elastic
Compute Cloud (EC2) service.
Tthe first company to lease VMs, the programmer-friendly EC2
Web services API and their pay-as-you-go pricing popularized
the “Infrastructure as a Service” (IaaS) paradigm, which is now
closely related to the notion of a “cloud.”
Success of Amazon EC2 , several other IaaS cloud providers, or
public clouds, have emerged—such as Elastic- Hosts [2], GoGrid
[3], and FlexiScale [4]—that provide a publicly accessible
interface for purchasing and managing computing infrastructure
that isinstantiated as VMs running on the provider’s data center
27th Jan, 20202

3. IaaS Anatomy
• IaaS provider 5 characteristic
1. on-demand provisioning of computational resources
2. Virtualization technologies to lease resources
3. Provide public and simple remote interfaces to manage resources
4. use a pay-as-you-go cost model
5. “infinite capacity” or “unlimited elasticity”
• Private and Public difference
• Role of Virtualization
• Key of these characteristic
• Allocating resources efficiently
• Taking into account an organization’s goals
• Reacting to changes in the physical infrastructure
Cloud Computing - Part II 2
27th Jan, 20203

4. IaaS Anatomy
• Problems In VM Solutions
• Distributed management of virtual machines
• Reservation-based provisioning of virtualized resource
(Round robin, first fit)
• Provisioning to meet SLAcommitments
• RESERVOIR project
• Resources and Services Virtualization without Barriers
• Addressed above problems
Cloud Computing - Part II 3
27th Jan, 20204

5. DISTRIBUTED MANAGEMENT
OF VIRTUAL
INFRASTRUCTURE
Cloud Computing - Part II 4
27th Jan, 20205

6. Distributed Management
* Manage the virtual infrastructures themselves
• Efficiently selecting or scheduling computational resources
• VM-based resource scheduling
• Static approach (initially selected as greedy strategies)
*Efficiency approach (VI managers must be able to support flexible
and complex scheduling policies and must leverage the ability of VMs to
suspend, resume, and migrate.
• Solution
• Virtual Infrastructure Manager
• Managing VMs in a pool of distributed physical resources
• Case Study
• OpenNebula
Cloud Computing - Part II 5
27th Jan, 20206

7. Reservation-Based Provisioning of Virtualized
Resources
Demand for resources
Provisioning to Meet SLA Commitments
IaaS clouds can be used to deploy services that will be consumed by
users other than the one that deployed the services. For example, a
company might depend on an IaaS cloud provider to deploy three-tier
applications (Web front-end, application server, and database server) for
its customers. service owners will enter into service-level agreements
(SLAs) with their end users, covering guarantees such as the timeliness
with which these services will respond.
27th Jan, 20207

8. DISTRIBUTED MANAGEMENT OF VIRTUAL
INFRASTRUCTURES
Managing VMs in a pool of distributed physical resources is
a key concern in
IaaS clouds, requiring the use of a virtual infrastructure
manager.
OpenNebula is capable of managing groups of
interconnected VMs—with support for the Xen, KVM, and
VMWare platforms—within data centers and private clouds
that involve alarge amount of virtual and physical servers.
OpenNebula can also be used to build hybrid clouds by
interfacing with remote cloud sites
27th Jan, 20208

9. VM Model and Life Cycle(OpenNebula)
• VM model attributes
• A capacity in terms of memory andCPU
• A set of NICs attached to one or more virtualnetworks
• A set of disk images
• A state file (optional) or recoveryfile
Cloud Computing - Part II 6
27th Jan, 20209

10. VM Life Cycle(OpenNebula)
1. Resource Selection (Rank scheduling)
2. Resource PreparationThe disk images of the VM are transferred to the
target physical resource.
• Contextualization
3.VM Creation-booted by the resource hyper reseources
4. VM Migration-The VM potentially gets migrated to a more suitable
resource (e.g., to optimize the power consumption of the physical resources).
5.VM Termination. When the VM is going to shut down, OpenNebula can
transfer back its disk images to a known location. This way, changes in the
VM can be kept for a future use.
27th Jan, 202010

11. VM Management
(OpenNebula)
• Management Areas
• Virtualization
• physical resource-such as Xen, KVM, or VMWare, to control
• (e.g., boot, stop, or shutdown) the VM;
Image management-by transferring the VM images from
an image repository to the selected resource and by creating on-the-fly
temporary images;
Networking- networking by creating local area networks (LAN)
to interconnect the VMs and tracking the MAC addresses leased in each
network.
Cloud Computing - Part II 7
27th Jan, 202011

12. Virtualization
• How?
• Interfacing with the physical resource virtualization technology
(hypervisors like Xen, KVM)
• More detail
• Pluggable drivers
• Decouple the managing process from the underlying technology
• High-level command
• start VM, stop VM
• Driver-based architecture
• Adding support VIMs by writing drivers
Cloud Computing - Part II 8
27th Jan, 202012

13. Image Management
• How?
• Transferring the VM images from an image repository to the
selected resource and by creating on-the-fly temporaryimages
• More detail
• What is image?
• Virtual disk contains the OS and other additional software
• Image management model
Cloud Computing - Part II 9
27th Jan, 202013

14. Networking
• How?
• creating local area networks (LAN) to interconnect the VMs and
tracking the MAC addresses leased in each network.
• More detail
• virtual application network (VAN)
• the primary link between VMs
• OpenNebula dynamically creates VANs
• physical cluster
• set of hosts with one or more network interfaces
• each of them connected to a different physical network
• Networking Model
Cloud Computing - Part II 11
27th Jan, 202014

15. Presented by Majid Hajibaba
Networking Model(OpenNebula)
Cloud Computing - Part II 12
27th Jan, 202015

16. SCHEDULING TECHNIQUES
FOR ADVANCERESERVATION
OF CAPACITY
Cloud Computing - Part II 13
27th Jan, 202016

17. Advance Reservation
• Demand for resources is known beforehand
• Example
• an experiment depending on some complex piece of equipment is
going to run from 2 pm to 4 pm
• Commercial Providers
• Infinite capacity
• Private clouds
• Finite capacity
• Reservation lead resource to be underutilized
• Haizea
• Lease manager
• Scheduling backend by openNebula to support provisioning models
Cloud Computing - Part II 14
27th Jan, 202017

18. Reservation with VMs
Virtual machines are also an appealing vehicle for
implementing efficient reservation of resources due
to their ability to be suspended, potentially migrated,
and resumed without modifying any of the
applications running inside the VM. However, virtual
machines also raise additional challenges related to
the overhead of using VMs.
•Preparation overhead- the disk inage can be fly or
transferred to the pysical node where it is needed.
•Runtime overhead
•Haizea- Advanced reservation leases, where the resources must be
available at a specific time. Best-effort leases, where resources are
provisioned as soon as possible and requests are placed on a queue
if necessary. Immediate leases, where resources are provisioned27th Jan, 202018

19. Cloud Computing - Part II 15
27th Jan, 2020
Leasing Model
We define a lease as “a negotiated and renegotiable agreement
between a
resource provider and a resource consumer, where the former agrees
to make a set of resources available to the latter, based on a set of
lease terms presented by the resource consumer.”
The following availability terms
• Start time may be unspecified (a best-effort lease) or specified (an
advance
reservation lease). In the later case, the user may specify either a
specific
start time or a time period during which the lease start may occur.
• Maximum duration refers to the total maximum amount of time that
the
leased resources will be available.19

20. Reservation with VMs
• Challenges
• Preparation overhead
• Runtime overhead
• Haizea
• Leases
• Advance reservation
• Best-effort
• Immediate
Cloud Computing - Part II 16
27th Jan, 202020

21. Haizea Lease Scheduling
• Backfilling
• How to address preparation and runtime Overhead?
• Disk image transfer before start
• Caching
• How does best-effort lease?
• Scheduling using queue
• Backfilling algorithm
• Depend on required disk image
• VM suspension/resumption
• How does advance reservation lease?
• EDF algorithm for preparation overhead
• Without preemption for Runtime overhead
• Pluggable policy
• Combine best-effort and advance reservation
• Overcome utilization problems
Cloud Computing - Part II 17
27th Jan, 202021