Cloud Computing Virtualization


hi so welcome to our ah next lecture on cloud
computing we will ah quickly go through the virtualization concept we have already discussed
some of these aspects in our previous lectures but i thought that we will have some few more
slides on it ah one of the core concept so virtualization is one of the core ah concept
or one of the core prime mover or having these cloud computing for which the cloud computing
is existing today and if you see this virtualization is not a new concept it is already there we
many of our are used to virtualize we virtualize a a linux system over a window system a window
system over a linux system so i have a some sort of a virtual realization of the things
right and if you look at that other end we are used
to different other resources like networking etcetera so i have a we can have a virtual
lan or v lans which is pretty popular which we have ppn virtual private networks and so
and so forth so virtualization is there and this cloud computing architecture technology
tries to exploit this a feature for giving different services of cloud so as as we have
seen this iaas or infrastructure has a service so from the subscriber point of view what
what the subscriber gets so access to virtual computers right network accessible virtual
storage network infrastructure components like firewall configuration services etcetera
so as as if i i being a subscriber i get access to a virtual machine right or set of machines
and then i may want that particular machine ah with particular storage and other configurations
so i dont know where it is how it is configured but for me if through this particular that
interface it is it is a type of machines which i am looking for even it is it is possible
that i i have a combination of this virtual machines along with a particular backbone
network and realize a a network infrastructure for my purpose right so this is this is that
what subscriber gets for me and how uses and what i pay for it is basically typically per
cpu hour gb data stored may be on a hourly basis network bandwidth consumed on a particular
rate network infrastructure used that how much ip address which are the routers etcetera
and value added there can be value added services like i monitoring automatic scaling and type
of things are the things which are value added services
so providers theoretically have a large pool of virtual machines that means if i am a cloud
service provider i say that these are my flavors of virtual machines which are i am having
at my thing right ah and then i i say that if you want a set of virtual machine then
i go on ah provisioning it right and usually these are into different categories though
theoretically i can configure any of the virtual machine but in practical sense it is it is
under with specific configuration and etcetera like i say that i have vm with very low things
like in case of a incase of our meghamala iit kharagpur we have a virtual three three
type of virtual machines one is iit kgp small vms which has a four gb ram with twenty gb
ah hard disk space and these are the other processing etcetera we have a iit kgp large
which has a eight gb ram and so on and so forth which have a we we have a iit kgp extra
large which has a sixty gb ram say these are the three flavor
now based on that based on my backbone resource availability i have a mix and match of the
things not only that the type of vms i will have is also dependent on type of request
i get on this vms right if i have a heavy request on only the smaller type of vms then
i want to provision that smaller type of vm much more than the higher things and so and
so forth so these are the considerations which come into play right so like in this case
a has a access to vm one to vm two right whereas b has a access to vm three where the client
c has a access to v m four to v m six where the providers retains vm seven through vmn
for its other users so that can be that can be ah a modeled and go on doing that and this
these are typically done like that so ah so if you look at the iaas component stack and
scope of control iaas components stack comprises of hardware operating system middleware and
applications layer so these are the typical thing so operating system is built into two
layers lower the most privilege layer so if we look at the more code to the operating
system is occupied by the virtual machine monitor or vmm which is also called the hypervisor
right higher layer is occupied by the operating system running within the vm called a guest
operating system so the as we have seen that ah we have a like
in this case in the middle if you see that i have the bare metal hardware where the cloud
provider have a total control cloud subscriber has practically low control over that i we
have a hypervisor right or vmm or virtual machine monitor where the administrative control
of the cloud provider are cloud subscriber can basically request a request for vm etcetera
to this hypervisor and then ah we have other layers like guest os middle ware like java
etcetera application like mail and crm and other type of things where the subscriber
has the total control in case of a iaas where the provider does not have any control or
does not provider does not control it ah for the thing right so this is the way it goes
on so if we ah if we look at it the type of control at the upper layers are moved to the
subscriber and type of control at the lower layers are moved to the provider side
so in iaas cloud provider maintains total control over the physical hardware and administrative
control over the hypervisor layer subscriber control the guest os middleware and the application
layers which are which over the subscriber subscriber is free using providers utility
to load any support operating system ah things so if the if the provider ah provides the
support the subscriber can load any guest os subscriber typically maintains complete
control over the operation of the guest operating system in each vm so though the subscriber
once it loads it has the total control over the guest os like if i if if the cloud provider
is providing me some linux os and hypervisor an allows me to load linux or windows etcetera
so if i load some flavor of linux then i have a total control over that particular thing
iaas components stack and other thing the hypervisor uses hardware to synthesis one
or more virtual machines or vms is an efficient isolated duplicate real machine subscriber
rents access to vm the vm appears to be subscriber as actual computer hardware that can be administered
that is powered on off peripherals configures via commands send over the network etcetera
so that it it because what i am having as a subscriber is primarily interface with the
provider and i can go on communicating with that
so typically if you look at a typical architecture so there are several components over at the
top is the cloud manager there are different cluster managers which are which are at the
second level of the things and below that is the computer ah manager or cms right and
the cm is basically at the has this hypervisors and different type of virtual machine immolated
over the things right so we have at the cloud manager and the data object storage which
is the muster data base of keeping track of the things then at the lower level we have
a persistent local storage like so that means hm the storage which does not go have a when
the cloud provider is not there when the vms are ah not used or shut down and type of things
so if we look at that three level hierarchy of the components h have one is the top level
is responsible for the central control like the cloud manager the middle level is responsible
for management of the management of possibly large computer cluster and may be geographically
distance from one another so if you look at this cluster manager it is it is it is managing
this large clusters which maybe geographically spread and the third is the bottom is responsible
for running the host computing system on which virtual machines are created so these are
running this host computing system where the vms are created right so this is typically
three layer of control subscriber quarries and comment generally flow into the system
at the top are forwarded down through the layers which either answer quarries or execute
commands so iaas cloud architecture ah cloud manager
can be geographically distributed within a cluster manager computers manager is connected
by a high speed network so this is this ah if you look at the cluster manager these are
all connected through a high speed ah network so now if you look at the quickly that operation
of the different things so what we have cloud manager at the top then the cluster manager
then the computer manager which manages the individual things right so if you look at
the cloud manager duty is the public access point of the cloud that is the first thing
so whenever somebody access the cloud in the public where the subscriber sign in for accounts
manage and resources they rent from the cloud etcetera cloud manager has the mechanism authenticating
the subscriber whether the authentication mechanisms generating or validating access
credential with the subscriber when the communication with the vms like when it it is basically
the of the whole system top level resource management so for subscriber request cloud
manager determines if the if the cloud has enough free resource or to satisfy the thing
so subscriber resource on the things so cloud manager maintains a meta data information
so if i request for a particular vm or a set of vms which are not available on the cloud
then it becomes a ah the cloud manager has to take a call so it is a it is it manages
at the subscriber level there is a thing called data object storage or dos dos generally store
the subscriber metadata as we are talking about the user credential operating system
etcetera dos devices usually single ah is single for a cloud
so what we say that the particular dos service is for a particular cloud so it a it maintains
a ah what we say registry or a cataloging or a metadata service for that whole cloud
so it is the binding block of the ah of the whole thing that what are the services available
whether the free resources available etcetera are all updated in the those so if we look
at the cluster manager that this ah middle level so it is responsible for operation of
collection of computers that are connected via high speed local area network right so
that is the ah it manages the lower level computers cluster manager receive resources
allocation commands and quarry from the cloud manager at the top and calculate whether the
part or all of a command can be satisfied using the resource of computers in the cluster
that means ah when the cluster manager gets a request it checks that whether the its available
resource with within the cluster whether it is able to satisfy the things and accordingly
it says a appropriate signal to the things like so that it can accept or cannot accept
and type of things cluster manager quarries the computer manager for the computers in
the cluster to determine resource availability returns message to the cloud manager that
mean it it it becomes a it ah middle ware or a agent between the cluster manager and
the cloud manager if the if you look at the operation there it get the direction for the
cloud manager and then instructs the computer managers to perform resource allocation reconfiguration
de allocation of resources and etcetera etcetera is cluster managers connected to a persistent
local storage that means it is the so called coat uncoat ah non volatile storage when the
ah particular virtual machine is ah shut down deprovision or some issues come ups so that
it has a a local storage into the things so the pls provide persistent ah disk like
storage to the virtual machine ok so it is as if persistent disk like the same things
are there so the data when you next time log in it will be available with you so look at
the if we look at the operation of the cloud manager ah at the lowest level in the hierarchy
cloud manager runs on each computer system uses concept of virtualization to provide
virtual machine to the subscriber so cloud manager is basically providing the ah vms
so it is the to the subscriber this providing vms it is it is responsible for that computer
manager maintains status information including how many virtual machines are running etcetera
etcetera so it maintains the status information as of now how many virtual machines are running
also cloud manager uses command interface to hypervisor to start stop a stop suspend
reconfigure virtual machines so it also same the if there is a need to the ah computer
manager so all this what we see is virtualization place important role right so how this virtualization
is made and what are the aspects etcetera thing we will try to see quickly in the subsequent
slides so virtualization is a broad term it can be a virtual memory it can be virtual
network it can be virtual storage etcetera so anything which can be virtualized is a
virtualization aspect our primary focus is the platform for virtualization right virtualization
basically allows one computer to do the job of multiple computers right by sharing resources
of a single hardware across multiple environment so that is important so that means i have
a bare metal then i create different machines or virtual machines which can cater to the
things but at the at the back end my same bare metal is running now this this becomes
a very tricky so because suppose you are running if you are having the your bare metal on particular
ah environment and then you are running one machine on windows one machine guest os on
linux one machine or some other things etcetera etcetera then there is a problem of issue
of how this ah say ah instruction set of this guest os will be running on the hardware at
the end so that there are issues of like that there are issues of application sizing there
are issues when the vm says that i require more resources or i want to release other
things how it manage and so and so forth so of we look at the hardware so if it is a no
virtualize system a single os controls the all hardware platform resources if it is a
virtualization system it takes the possibly ah it makes it possible to run multiple virtual
containers on a single physical platforms so i can have multiple virtual containers
which which can run on or which can be plugged into a single physical ah infrastructure so
ah we are we are somewhat experienced with this virtualization so virtualization is a
way to run multiple operating system and ah user application on the same hardware so that
is virtualize so i can i can have two different operating system on the same that means i
say i virtualized one this os out of this right
so run both windows linux on the left of so how is it different for dual boot dual boot
also we are doing that once windows etcetera here in case of virtualization this ah different
os or all the os s are running ah together right or simultaneously the oss are completely
isolated from each other right they are completely isolated from each other ah in case of a true
virtualization now another ah important aspect of things which come into play is the virtual
machine monitor or hypervisor so a hypervisor or a virtual machine monitor runs the guest
os directly on the cpu right so that means i have a guest os and i am i at the at the
ah as my which has been immolated or installed by the ah subscriber or running on the or
its application on the of the client and then the hypervisor is responsible to execute this
or ah run this guest os directly on the hardware at the back and the things so this works if
the guest os uses the same instruction set of the os host os right if it is a different
instruction set then there is a instruction translation should come into play right so
there are several issues as as depicted by popek and goldberg that three goals of virtual
machine architecture are there one is that equivalence the vm should be indistinguishable
from the underlying hardware so as if the virtual machine is running on the hardware
itself so what we say it is equivalence or resource control the vm should be incomplete
control of the virtualize resource so if you give me four gb machine thirty gb
sixty gb hard disk and so and so forth that i should have to complete control over the
thing right as a subscriber and efficiency most vm instruction should be executed directly
ah on the underlying cpu without involving the hypervisor so that that is another thing
so efficiency will increase if the most of the virtual machine instruction should able
to execute directly on the underlying cpu without intervention of this other involvement
the hypervisor so these are the aspects which need to be looked into when we have virtual
machine monitor or hypervisor influence which allows me to emulate vms so again in the same
ah work these popek and goldberg describe and also give a provided a formal proof of
the requirement of the cpus instruction set to allow these properties to happen like to
allow truly virtualization happen at the iaas level the main idea is classifying these instruction
into three different thing one is privileged instruction which ah which cause a trap in
the executed in the user mode there is a sensitive instruction which ah change the underlining
resources that is doing io ah or changing the fact that the guest os is running on the
bare metal etcetera a sorry ah means first of all changing the io or the page page tables
or observe the information indication on the current privilege level so if you if you remember
you are operating ah if you are basic architecture things so i have different level of operations
like level zero level one and so and so forth so more the lower the level as i go more on
the more closer to the bare metal so it all depends that your virtualization at which
point it operates more the higher level then more latency will come into play more translation
will come into play so we need to look into the things that what level operations i need
to do specially when we do iaas type of ah operations
and the former class of sensitivity instructions also called controls sensitive or latter call
are behavior sensitive operations right so what ah this popek and goldberg so that if
we can run a virtual machine with all the three desired property if the sensitive instruction
are a subset of a privilege instruction in their work they have shown that ah this sort
of things are true and we can realize a true virtual machine so the same thing ah or thus
whatever we are discussing so we have this vms and this hardware over this this virtual
machine monitors is running and its allows us to emulate this different ah vms right
so it can be your ah so some where other this whatever the guest os running on the vm the
instruction set need to be converted which can be understood by this hardware right so
that is the our bottom line so whether it is whether the whole vmm will have a total
abstraction or also have some sort of ah overlapping with the things ah that that depends on whatever
the implementation so if we look at the approaches typical approaches
to virtualization so if you see the evolution of this virtualization one is what we say
first general a first generation or full virtualization so binary rewriting so it is software base
vmware and microsoft supports this so whatever the virtual machines are generating a dynamic
translator rewrites to the ha underlining hardware and so and so forth in case of a
second generation virtualization or para virtualization cooperative virtualization so it is a cooperative
virtualization the guest os or the is modified and vmware xen as it so it basically if you
look at here so it penetrates little bit into the vm that means that is modified on the
things in case of a third generation or what we say silicon based that is hardware assisted
virtualization unmodified guest vmware and xen on the virtual aware ah on the virtualization
aware hardware platform so now your hardware platform is a virtualization aware so it is
hardwared hardware assisted virtualization so in this cases this hardware platforms are
not aware of this virtualization whatever is done done by this your hypervisor so there
are ah different things in in case of full virtualization for generation offering so
dynamic binary translation of the source code as we have seen all the hardware is emulated
including the cpu to popular open source emulator many of you might have seen qemu and bochs
these are the two popular emulator which ah change that open source full virtualization
there are advantages of emulation layer like you dont bothered about that two things are
separated ah total vm portability like i can put take the vm from one to another so put
total portability because only the vmm i need to understand those there are drawbacks this
hardware emulations comes with a performance vice whenever you want to emulate a hardware
etcetera performance vice in traditional x eight six architecture os kernels expect to
run on the privilege ring zero and now you want to run them at a higher level and that
that need to be there should be more latency and other things come into play
so we need to pay for that or what we need to there is a performance price for that things
in case of para virtualization guest os is modified and thus run kernel level operating
system at ring one or ring three that is much higher level guest is fully aware of how to
process privileged instruction privilege instruction translated to vmm is no longer necessary guest
operating systems uses specialized api to talk to the vmm right so it is para that means
your vmm is penetrating to use some sort of a some into the ah ah your ah vms so that
now it is not full translation but i have a para virtualization the performance increases
and this things is becoming pretty popular so in case of para virtualization recompiling
of the os kernel is required because there is penetrating into the things installing
of para virtualized drivers are required so that it it takes care at your guest os level
the third one is a hardware os running on a ring zero vmm uses processor extensions
like in case of intel vt or amd v and type of things to interpret the privileged things
hardware assisted virtualization ah removes many problems that take place in writing a
vmm into a writing a vmm into a rather it is a big challenge so that it is hardware
assistant so that means you go directly over thing is totally fully compiled in doing so
you lose that portability of the vms right it becomes a more tricky issue because you
need to migrate to something which understand that hardware pros it allows run unmodified
os right you the oss need to not be modified cons is the speed and flexibility is issues
that when that your flexibility you lose and the speed it depends when the migration etcetera
comes into play so with that we will we will just touch upon some of the one more aspects
of the things what we say network level virtualization so that now what we are tooking as the hardware
now we are looking at the network level that means i need to emulate a network over a virtual
network over a given network right so typical approaches what we have ah seen is that ah
in case of a network technology that is whether ip base or atm base layer which layer of virtualization
is there there are architectural domain like network resource managements spawning network
and type of things level of virtualization which is node level or full virtualization
of the things like ah if you look at suppose i have i have a network like this then i can
emulate a virtual network like this or i can emulate a virtualize network like this right
so this is my base network so it i can have two type of virtual network on the thing over
and above of this thing so i emulate ah a ah virtual network why to virtualize now ah
what people say that ah not only internet in your own networking paradigm is becoming
paralyze like you dont have much in go space hard to come with a one size fit all type
of architecture in number of cases like you have a network configuration and fits everybody
so why not to create something that all sizes fit to one instead type of scenario so that
is the ah things what we want to do so testbed for future and there is that is and any futuristic
working on the network protocols and etcetera require a testbeds for that putting a whole
networking things may be a challenge so i can have a ah sort of a virtualize network
and ah this is not suddenly came into play we have lot of things which are already in
play that is a virtual private network active and programmable networks overlay networks
these are already in place for ah several years and this is being emulated over that
and if you look at there are different type of aspect like business model architecture
design principle design goals and those are the things and there are different aspects
of those ah like infrastructure providers service providers end user brokers these are
the different players and they have a very ah complex relationship between ah between
each other that how things will be there and this business model is not only true for network
virtualization it is true for other sort of virtualization as well
and if we look at that architecture so if i have that infrastructure provider then i
have a service provider one which is have some sort of a virtual network emulated from
this like making active some of the networks etcetera and i can have other type of service
provider two sp two emulating other type of networks as well so i can emulate different
variety of network on a basic underlying network so i can realize different networks into the
things so so these are also becoming very popular so having say not only having servers
like i want to conduct a particular testing over across the ah across ah across the geographical
space so not only i want this servers into the ah place i also have a particular network
infrastructure into play right then those ah then if you want to have those type of
things then i not only require the virtualization at the iaas level or the server level virtualization
i also require a network level virtualization into play and making this overall i can have
a virtualized ip infrastructure into place right so we will continue our ah discussion
in in in our subsequent lectures ah we will stop here today
thank you

Daniel Ostrander

Related Posts

Leave a Reply

Your email address will not be published. Required fields are marked *