Mobile Cloud Computing -II

Mobile Cloud Computing -II

Articles, Blog , , , 1 Comment

Hello. We will continue our discussion on
Cloud Computing; rather we will continue our discussion on Mobile Cloud Computing. So,
what we are discussing about mobile cloud computing said that the mobile is a mobile
devices these days are omnipresent right. There is a huge glorification of smart mobile
devices which has fairly good capability of doing lot of a applications or running lot
of applications. As there is a as the capability increases, the need for running more computational
and data intensive applications also there is a increase in the those type of applications.
Now, in though there is a increase in this resources of the mobile devices, but what
we still feel that some of the cases had it been a background, backend, some more computing
services that could have been better like we execute some part here, some part offload,
some part of the computation or data on the backend devices where what we have seen that
cloud may play a important role. But there are several issues into the things right that
any transfer of data and processes involves costing. Costing in terms of execution time
specifically that if the delay is more than something, it is it becomes unviable also
we see that is a question of reliability when once your application runs goes to another
place to run. So, if there is there are a lot of dependency
involves like intermediate network or the wireless network, the availability of the
resources at the other end or failure or fault in the other end and of course, the fault
in the device at the a mobile device itself. So, whereas, when you offload all those things
come into play right then we have also seen there are issues of security right. So, far
the application running your own device, you have someone on the process and the data over
it right, but whenever the application running in some other premises or some other devices
or cloud, which is maintained and owned by some third party, then your there is a concerned
about privacy and other security issues the regarding those data and a processes.
So, this becomes pretty tricky issue and nevertheless mobile cloud computing is becoming pretty
popular because of several reasons. First of all you can now have think of a large application
running on or rather controlled by a small devices, there can be things that the your
process may be using somebody elses data. So, you run somewhere where the data is there
and the process is there or the mobile device acts as a collection of data which is contributing
to some central repository, some process at this device or some other device can do. So,
there is there are issues of in exchanging of data interoperability issue and sharing
of this sort of devices right. Or what we see a evolution of this that internet
of devices come into play where cloud plays as a central role of managing the so right.
So, we will look at a some of the aspects today regarding a mobile cloud computing in
continuous in our with our previous lecture of mobile cloud computing, and rather we will
try to refer some of the research works which will give us a opportunity or give us a more
better view of how what this what are the different challenges and how to handle them
and so and so forth. So, we will look at a mobile cloud computing,
as we have seen some of the key challenges are that that requires dynamic partition of
a application to optimize energy saving execution side right. So, two things as we mentioned
earlier also, one is energy saving and to optimize the execution time right. If the
time is beyond a limit then there may be that then that process may not be viable to run
on the things also requires software or something middleware that decides that the app launch
which parts of the application will be executed locally and which part of the application
need to be executed on the externally or remotely or in the cloud right.
So, it is a classic optimization problem right. So, we can do this sort of means dividing
this application into different parts, some of the parts or the component of the application
need to be run locally and some of the remotely, this can be done statically like I have a
priori knowledge of this sort of things can be done or in some cases I may not know the
a priori that which need to be locally or locally or remotely, primarily it is if it
is dependent on the data and other type of things right. So, what will happen I need
to have a dynamic these and all the things. And as we understand whenever we need this
type of thing synchronize a sense scheduling and the resource management come into play
in a big way right. So, all those things here and when you are specifically doing the devices
and say infrastructure, which are owned and maintained by different authorities then it
is a more tricky. So, like mobile devices is something and cloud is maintained by other
third party, intermediate you have service provider we are who are again have a different
type of authority. So, all those things makes a very complicated issue need to be solved. So, this is one of the work is as referred
in the in this paper, there is mobile assistance using infrastructure right. So, it is one
form of mcc. So, MAUI enables the programmer to produce an initial partition of the program.
So, programmer marks the each method as remoteable or not that means, whether a remote or local,
native methods cannot be remoteable which are need to be done on the on the device itself
MAU framework uses the annotate to decide whether the method should be executed on cloud
server to save energy and time to execute the thing right.
So, there are a couple of times the time to execute on the cloud there is a time to transfer
this process and data to the cloud and get back the data if as if the need demands and
MAU server is a cloud component. So, there is a MAU client which is on the mobile side
there is a MAU server which is a cloud component, the framework has necessary software modules
required for the workflow. So, that there is a client server type of architecture where
what they propose and can be able to synchronize. So, if you see; there is the app and there
is a profiler and the solver profiler basically profiling of the application and profile along
with the server partitions the application these gates executed here, and with RPC calls
and then it in a two and four things Lan server communicate to the profile as sync among itself
when the apps run in a some part on the app and some part on the some part on the device
and some or what we say smart mobile divider, SMD and some part on the cloud right. So, there is another work which is what is
called comet quad, code offloading by migrating execution transparently right. So, works on
unmodified application no source code required that is on the app itself. Allows threads
to migrate between the machine depending on workload, they based on the things that threads
can migrate, it implement a distributed memory shared memory or DSM model to runtime engine
right those who have gone through computer architecture or advanced computer architecture
and various type of things, they understand that this DSM model. A DSM allows transparently
movement of the threads across machine. In computer architecture DSM is a form of memory
architecture where physically separate memories can address as one logically shared address
space, there is the core of the or there is what we say very quick view on the what this
DSM says. So, likewise there is a phone OS and remote
OS. So, it need to maintain this memory states and there is a distributed memory synchronization
and the executed this processes other codes are executed or offloaded by migrating execution
transparently. So, it is not that the neither the source code is required nor that has nor
the a user end there is known that this things are going on at the backend. So, what it requires
only program binaries right execute multi thread programs correctly. So, it does not
require source code only the binaries; furthermore improvement on the data traffic during migration
is also possible by sending only parts of the heap that has been modified. So, it is not transmitting the whole heap,
but the part of the heap which has been modified. So, what we see the key problem to address
or solve is a at the at its core MCC framework, must solve how to partition a program right
or execution on heterogeneous sources. Thus not only partitioning, it need to be executed
on heterogeneous resources right it one on the mobile device side which may be some mobile
related OS when you are running on the cloud side that is some other kind of environment.
So, it is a heterogeneous environment how things need to be run need to be looked into.
So, it is a task partitioning problem as we have seen in other computer architecture and
organizations, widely studied in processor scheduling as a job scheduling problem also
right we do have this sort of challenges out here. So, need to be looked into from that
angle. So, task partitioning problems. So, what we have typically in a task partitioning
problem in mcc a call graph representing the applications method or call sequence right. So, it is a call graph, attributes of each
node in the graph in denotes a energy consumed to execute the method on the mobile device
and energy consumed to transfer the program states to the remote server. So, one is the
energy consumed in the mobile device and transferred the things.
What we are considering that the other end that is the cloud end, the it is pretty resourceful
and efficient. So, the energy consumes in is in executing that in the cloud end is negligible
in compared to the energy consumption at this end at the device end. Rather the device does
not much clear about that, but you need to clear about that executing its own locally
the portion to be executed, and the transfer cost of this the same memory state and processes
and so and so forth as things required right and data memory state and so on. So, output
what we get as a partition the methods into two sets one set of the matter executes at
the mobile device, and the second state set it goes to the execute on the cloud.
So, go if we look at the goal and constraint, energy consume must be minimized. So, I need
to minimize the energy consumption, there is a limit on the execution time of the application.
So, it should within the threshold say time capital T, other constant could be some methods
must executed on the mobile device total monetary cost etcetera. So, two we have couple of a
constant. That means, we need to goal is to reduce the energy consumption, time should
be within that particular threshold value. And there can be other constant like some
of the application some part of the application may need to be executed in the device itself
like I may not be able to offload it. And I may look at the total costing of the
thing the total cost should not execute some something some portion more than that, so
these things are necessary. So, if we look at that little mathematically.
So, we have a execution graph right out of that what we say that these are v1, v4, v9
in this case are native task; that means, needs to be executed in the device itself
right I cannot offload it. So, there can be so, highlighted nodes here must be executed
device, edges represent the sequence of execution that how the sequence of execution will be
there. Any non highlighted node can be executed either locally or on the mobile device right
can be locally or on the mobile device. So, if it is offloaded to the local device,
this first four senses that how much energy it could have saved right or in other sense
had it been worked on the local how much energy could have been taken. So, that has been saved
if I offload it to the remote, but I need to also look at that where the costing of
transfer this data right. So, there is a cost involved or there is a energy cost involved
energy expenditure involved in transferring the data from the device to the remote device
right that has to be incurred by the mobile device right. So, that are needs to be there.
So, if we need to maximize this. So, if it is if this result is a positive or the maximum
of the maximum if you can maximize, then I have a substantial energy saved otherwise
cannot be there the there is no benefit of doing that right and of course, it should
be within a total execution latency should be less than L right we should not have this
whatever we do need to be less than the particular latency L right. So, that it should be within
that particular things and I need to look at that a dependency also like execution of
one need to be executed, if there is a dependency in the graph that need to be executed out
here. So, as we are discussing there are two type
of partitioning one is static and dynamic as we have seen. Static in case of a static
when the application is launched the invoke a ILP solver and which tells where each method
should be executed. So, it says that statically says that when it is launched. So, there is
there are also heuristic to find a solution first term that there are we can deploy. In
case of a dynamic or adaptive partitioning, what we do for long running program environmental
conditions may vary right if it is a executing for a long time environmental condition may
vary like a requirement of the data and even the availability of the remote resources transferring
bandwidth all those things can vary. Depending on the input the energy consumption
of a method may vary so; that means, in other sense I need to take a call that which partition
need to be executed locally and remotely because of that varying nature of the things there
are hosting may go on changing right. So, that is another challenge out here. So, there are as we see there are when we
do this type of partitioning and transfer the job executes something locally, something
remotely or everything within a time limit, a particular threshold values, threshold time
and we want to maximize the energy saving right.
So, these are the different constraint and in order to achieve this sort of issues. So,
there are several challenges and issues need to be addressed. So, mobile communication
issues that is low bandwidth one of the biggest issue, because of the radio resource of the
wireless network is much more scarce than the wired networks. So, the it is the availability
band availability of this bandwidth mobile bandwidth or wireless bandwidth at times is
pretty less in compared to a wired type of things or. So, that is not only the less it
also varies on if it is a long running execute executive means long running process then
it may vary over time. So, getting a guaranteed bandwidth is a challenge
service availability mobile user may not be able to connect to the cloud to obtain a service
due to traffic congestion network failure mobile signal strength and so and so forth.
Issue of a heterogeneity handling wireless connectivity with highly heterogeneous network
to satisfy mcc requirement always on a connectivity on demand scalability, energy efficiency is
a difficult problem. So, there are heterogeneous heterogeneity involved in it and there are
these are different challenges. So, these are more issues with the communication issues,
there are a few computation issues or computing offloading related issues; one of the this
is one of the major feature of the things like you are able to offload it.
Offloading it not always effective in saving energy as we have seen that if those parameters
are not made then it may not be that energy saving, may not be what we say within that
quote unquote profitable margin right. So, it may be costly stuff; it is critical to
determine whether to offload and which portion of the source code to offload. So, its it
is a critical challenge, especially as that there are different type of application with
different type of requirement etcetera and its it is a major challenge to decide that
where when to offload, how much to offload and so on and so forth. So, there are other trials or other type of
a initiative one is that code offloading using cloudlet right. So, if you look at cloudlet,
it is a trusted resource rich computer or cluster computers that is well connected to
the internet and is available for use by nearby mobile devices. So, it is a what we say we
can look at as a smaller version or of cloud, but it is resource rich, but it is more within
the locality of the mobile devices or nearby regions right.
So, offloading a code to the remote server and executing it, when you look at the court
code offloading part. So, this architecture decreases latency by using single hop network
and potentially lowers battery consumption using Wi-Fi, short range radio broadband wireless
etcetera. So, what we try to see that first of all of offloading and secondly, when we
make multiple hops then the cost increases. So, what you try to see that whether I can
have presence of smaller clouds, which can take care of this sort of application, this
sort of a executing this portion of the application. And if a single hop then we can have something
more or what we say that we can have more the better energy saving and better response
time, with having those resources much closer to the device. So, cloud code offloading using cloudlet.
So, we have a mobile devices, mobile networks and the remote cloud and there are if you
look at there are different small small different set of systems or mobile cloudlet, and cloudlet
nodes which takes care of these devices right or what we say that goal is to reduce the
latency, in reaching the cloud service servers use a servers that are closer to the mobile
devices. That means going for instead of the full place
cloud, we can have smaller version of the clouds or cloudlet. A cloudlet is a new architectural
element or that arises from the convergence of mobile computing and cloud computing right.
It represents a middle tire of a 3 tier architecture like mobile devices cloudlet and cloud. So,
that in other sense what we are trying to do in order to reduce the latency, and a better
energy savings possible better energy saving in one side the cloud is there, one side a
devices we made a we make a intermediate a thing what we say cloudlet right; so which
takes care of the things. Now when to offload is again a serious question. So, if we look at a very straightforward formulation
of the things. So, if we look at that equation that. So, what we say that Pc is the energy
cost per second when the mobile phone is doing computing; that means, mobile phone is computing
Pi is that energy cause the mobile phone is idle.
Ptr is the per second a cost per second when the mobile phone is transmitting data. So,
a it a mobile device can be either computing or idle or transmitting data. So, this three
component are Pc, Pi or Ptr or when energy when transmitting data. Now if you consider
c as the number of instruction, and m is the speed of the mobile compute the c instruction.
So, what we see that if we see that Pc C by M if it is offloaded Pc C by M minus Pi C
by S. So, what is s the speed of cloud to compute the C instruction.
So, if it is s is the p Ps s is the speed of the cloud then I we can see that idle Pi
is the idle time, C there is number of instruction by the speed. So, this P by C by S and the
transmitting P trd by B D means the total data to transmit and B is the bandwidth available.
So, if we offload. So, this is my overall Ballpark figure right this should be a positive
number. So, it says that I save energy right. So, this was my energy consumption if I offload
this consumption are not there and then then this is this is save of the saving of the
energy or if I say the server or the cloud server is f time faster, then I can look at
s equal to F into M right. So; that means, your speed of cloud to compute
C instruction and M is the speed of mobile phone to compute C instruction if it is f
times faster, and we can see that S equal to F into M right or in other we can rewrite
the program or rewrite that particular formula or substitute this s equal to F by M into
the formula to look to basically arrive this thing C by M, Pc Pi by F minus Ptr D by B.
This is a very simple arithmetic you just substitute and if you take appropriate common.
So, we come to a figure called in this sort of scenario right C by M equal to Pc minus
P i by F and minus tr. So, what we see from here that energy is saved
when formula produces a positive number definitely if it is a positive then only energy is saved
the formula is positive if D by A, D by B is sufficiently small compared to C by M right.
So, one side is D by B should be sufficiently small compared to C by M. So, that I have
a larger number on the left side. So, this is what we say that this is one of the condition
cloud computing. So, what cloud computing can potentially save energy from the mobile
devices, note that all application in the energy efficient when migrated to the cloud
are may not be energy efficient right. So, from this formula already say that first
of all D by B should be must lesser less than C by M and F should be sufficiently higher
which makes sense right. F is the how many times the cloud server is faster than the
mobile device; it is likely that the cloud server should be sufficiently higher otherwise
there is no point in offloading and getting executed at externally right if it is you
know. So, if that conditions are satisfied then what we see that we get a positive number,
and it is it is obvious that all the applications may not be always beneficial when you offload,
cloud computing service should be significantly different from the cloud services for desktop
because they must offer energy savings in this case.
The services should consider the energy overhead for privacy, security, reliability, data communication
etcetera which we have not directly considered in these cases. So, it is a ideally it should take care of
other type of features which in while calculating the things. So, if we look into different
other way of looking at us same a formulae formula is that offloading is beneficial when
large amount of computation C are needed with relatively small amount of data communication
right. So, we require large amount of computation
right with relatively less transmission of data right. Here also as we have seen that
C should be sufficiently large with D should be sufficiently small. Keeping like F must
be much faster and do we see that it is a thinks and we can see that in different scenario
how things will be there right. So, party computation overloading there are
several approaches, you can see several research papers also partitioning a program based on
estimation of energy consumption before execution. So, what we say that pre partition estimation
of the thing, optimal program partitioning for offloading is dynamically calculated based
on trade off between communication and computational cost at the at run time.
Offloading scheme based on profiling information about computation time data sharing at the
level of procedure calls right. So, at the level of procedure calls we can do. So, a
cost graph can be generated or constructed and branch and bound algorithm to applied
to minimize the total energy consumption of computation and total data requirement. So,
what we are looking at that it is not vanilla offloading some of the things to the things.
So, there are a the lot of calculation involved as we come back to the our previous premise
that one is that your energy savings should be higher, need to be maximized whereas, your
latency should be preserved. And we see that there are several other aspects
of the components are there like security, reliability, fault in the server system etcetera
and there are issues which are not directly under not and may not be a directly under
the control of the cloud, neither on the mobile devices which is what we say that intermediate
wireless network that also plays a things. So, that is there are issues of heterogeneity
need to be addressed and need to be looked into a in a totality it is not that isolated
I offload something, get some executed and type of things need to be looked as a in totalities.
So, that the user at the mobile end or the user get a seamless feeling or better resource
utilization or better energy savings at that things.
Thank you.

One thought on “Mobile Cloud Computing -II

Leave a Reply

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