YT YouTube2Text

YouTube2Text → Transcripts → Cloud, edge and fog computing for IoT: Part I

Cloud, edge and fog computing for IoT: Part I — Transcript

by NPTEL - Special Lecture Series · 11,499 words · 2,116 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:03so today it's going to be
0:04a very different session it's going to
0:07be
0:08in this completely different area of
0:11cloud edge and fog computing for iot
0:15and it's it's uh so cloud computing as
0:18we know it has already existed
0:20uh for quite some time now but uh
0:24recently there has been a lot of uh
0:27emphasis on uh edge computing
0:30and fog computing we'll see how these
0:32are
0:34similar but but not exactly the same
0:38and we'll see different flavors of edge
0:41computing
0:42also so that's uh going to be the
0:47broad topic that we'll discuss today
0:51um feel free to stop me anytime to
0:55you know if you have any questions
0:59okay so outline of today's uh talk uh
1:02we will kind of break it down into two
1:05for the two sessions we'll break it down
1:07into two
1:08uh parts uh so we'll start with cloud
1:11computing
1:13uh we'll go through uh the
1:16basics of cloud computing what some of
1:19the
1:19internal uh architectures
1:23um and and some of the challenges we
1:26face there so we will
1:27cover a lot of cloud computing aspects
1:30in the first session
1:32uh and we'll kind of motivate as to why
1:35cloud computing
1:37uh has issues uh when we look at
1:41internet of things and that's when the
1:46aspect of edge computing comes into
1:48picture
1:49and so in the second session
1:53uh we will look at uh edge computing
1:56uh some of the details of uh uh
1:59its characteristics uh their various
2:02uh uh flavors of edge computing that you
2:06can see that is
2:07fog computing cloud leds mobile edge
2:10computing
2:11and so on and uh we will uh then um you
2:14know
2:15kind of understand it a little bit
2:17better using some examples
2:19yeah uh i kind of um
2:22at the end of this i'll try to motivate
2:24you with some very
2:26uh interesting and very uh
2:30widely currently very hot areas of
2:33research in this
2:34uh in this topic so anybody who is
2:38interested in research also would be
2:40uh interested to hear that um
2:44so yeah so that's that's going to be the
2:46outline of today's talk
2:49so let me uh without any delay getting
2:52to what cloud computing looks like so
2:54i'm uh i'm assuming everyone has heard
2:58what cloud computing is right there is
3:00that is something that has been there
3:02for quite some time
3:04now uh and everybody would have at least
3:07heard of it if
3:08uh you don't know
3:11the exact details of how it operates
3:14uh so uh cloud computing uh
3:18there is an evolution and we will look
3:20at how we came to cloud computing
3:22uh there is a there is a very uh
3:26uh you know very well defined evolution
3:29of how
3:30uh cloud computing came about but
3:33to give you an introduction cloud
3:36computing as you can see
3:38is it it basically
3:41allows various users
3:45who are using a different kinds of
3:47devices let's say pc
3:50laptop smartphones or whatever it is
3:54to access some services
3:58or some resources uh via the internet
4:02uh and and they are try and they access
4:04it uh
4:06uh from uh from a remote uh location
4:09right and these remote locations are
4:11typically
4:12uh you call it cloud and they are
4:14actually data centers
4:16uh so that's that's how uh the
4:20uh cloud computing um
4:24basically the the whole uh flow of
4:28data happens there but formally if you
4:32define cloud computing
4:33it is it is a style of computing
4:37where the resources that you
4:41get from the cloud
4:45they are providers as services yes so
4:47there are different kinds of services
4:49that are provided by the cloud
4:51and these resources are these resources
4:54could be
4:55hardware resources or they could be even
4:59software services and
5:02if they are some hardware resources they
5:05are typically
5:06virtualized so if uh
5:10if you don't know what virtualization is
5:13it is nothing but uh there is a large
5:16pool of resources
5:18or hardware that is available and
5:22they are uh provided so there are
5:25physical resources available and they
5:27are grouped into logical
5:29uh resources and then every user gets
5:32up uh gets a logical uh set of resources
5:36um so so basically there there are the
5:38same set of physical resources which are
5:41shared among
5:42uh multiple users yes that's that's what
5:44uh
5:45virtualization achieves so
5:48uh uh the so this is a style of
5:52computing which
5:53uh which uses this virtualization
5:56uh technology and uh virtualization is
5:59one of the enabling technologies for
6:01uh cloud computing uh
6:04and and this virtualization is then used
6:06to uh
6:08provide resources in a shared manner uh
6:10to different users
6:12uh as as services over the internet and
6:15one of the interesting things about
6:17cloud computing is that uh you can
6:20dynamically scale uh the
6:24resources that are provided to a user
6:27so it's dynamically scalable so that's
6:30that's one
6:32interesting thing and you have to
6:34remember uh
6:35for the time being that this dynamically
6:37scalable
6:39uh part is uh something that
6:44is is valid until you know there are
6:48lots of users trying to access it
6:51that is human users
6:55now that also has come into question
6:59uh as we get into the iot era
7:02so that is one thing that needs to be
7:05kept in mind
7:06so there are uh there are these
7:10uh different resources that are
7:12available and
7:13uh and they are available in these
7:16so-called data
7:17centers and any user if he wants to
7:21access
7:22these resources or services
7:25they can access and there are particular
7:28ways of accessing it
7:30it has it's of course you have to pay
7:32[Music]
7:35and there there is a pricing mechanism
7:37that's associated with it for you to
7:39access
7:40so the advantage of cloud computing is
7:42that
7:44there is there is a very significant
7:48savings in terms of cost yeah because uh
7:52now the user or the let's say some
7:57organization uh does not have to set up
8:00those
8:01uh they don't have the startup uh
8:04uh cost so to say to set up the
8:08uh infrastructure uh to
8:11for computing yeah so they they can uh
8:16pay of course they have to pay something
8:17but they can pay
8:19uh and access those services from a
8:22third party provider uh and and that is
8:26much more cheaper than uh setting up
8:28your own uh
8:30infrastructure to do that so that's
8:32that's one of the advantages that's
8:34there
8:34uh for cloud computing um
8:37it is also pretty much known to be
8:41highly available that is
8:44there is there is a lot of redundancy
8:48mechanisms
8:50there are mechanisms that are there in
8:52place to
8:53make sure that the services are
8:55available
8:57around the clock and you don't face
9:00too much down time but it's it's
9:03something that is
9:05not entirely guaranteed so sometimes we
9:08we do hear that some of these
9:12cloud providers are down and then you
9:15you're not able to access some of these
9:16services
9:18from them right so that is that is
9:20possible but uh the
9:22one of the things that is uh
9:26tried to maintain in in cloud computing
9:29is that
9:30there is a high availability of
9:32resources and services
9:34and and of course scalability is the
9:37other
9:38advantage so now you have a large pool
9:41of resources
9:42so as an um as and when a
9:46user wants to uh
9:50you know derive you know more
9:52performance by
9:53accessing more resources uh the
9:57uh cloud provider can
10:00the there can be an automatic way of uh
10:03you know
10:03scaling the resources so that
10:06the user sees the improvement in the
10:09performance based on you know
10:10even the increase in the amount of
10:14workload that the user may want to
10:16execute
10:17so so uh these are uh
10:20some very uh you know well-known
10:24advantages of cloud computing
10:27so we will look at
10:31some of these aspects in terms of how
10:33these services are
10:35uh accessed and uh what are the
10:37resources that you access
10:39from the cloud so before getting into
10:42that
10:43uh as i mentioned there there is a very
10:45well defined evolution of how
10:48uh cloud computing came into existence
10:50right so
10:51uh here is this is what uh shows that
10:55um yeah so as you can see this first
11:00phase that was called the mainframe
11:02computing phase where
11:04every user had a so-called dump terminal
11:08so
11:09basically everything that every
11:12intelligent uh
11:13execution that needed to be done was uh
11:17sent to a mainframe computer uh which
11:20had a
11:21large amount of resources and you could
11:24execute
11:25uh your tasks in terms of
11:28jobs and once you have the result then
11:31it's sent back
11:32for consumption so uh that that was the
11:35initial way of
11:36how computing uh happened then
11:40uh the pcs or the personal
11:43computers started becoming more and more
11:46powerful
11:48and then that led to an era where
11:52it was no longer required as they became
11:55more and more powerful they became
11:57also equipped to handle uh
12:00tasks that uh required a lot of
12:02computation so you could do a lot of
12:04intelligent
12:05things on your personal computers itself
12:07so
12:08uh that's that's the second uh
12:12phase which is called the pc computing
12:14phase
12:15um but then that was uh that still did
12:18not
12:19uh exploit uh the uh
12:22the the full uh you know capability of
12:26how
12:26computing could uh be scaled
12:30so that's when uh some uh researchers
12:34thought of connecting
12:37uh your the personal
12:40computers laptops etc to
12:44uh to some other uh
12:47you know high-end machines or servers or
12:49other other pcs uh
12:52using local area networks so there there
12:55were
12:56local area networks that were used to
12:58connect
13:00these machines and that came to be known
13:03as the network computing
13:05phase um
13:08yeah so this this of course uh you know
13:11uh
13:11made it quite uh uh the the
13:15flick there was a lot of uh improvement
13:18in terms of what could be done
13:21uh how things could be scaled uh in
13:24terms of computing
13:25uh but uh you know we could go further
13:28right
13:29why why stop at one uh local area
13:32network
13:33so then people started connecting uh
13:38different local area networks yeah so
13:40multiple networks
13:42small networks in local
13:46areas started getting connected and that
13:49came
13:49to be known as the internet so now
13:53we you can see the full
13:56potential in terms of connecting things
13:59all over
14:00the world so that that that phase was
14:03called the internet computing phase
14:06um but still you know there were uh
14:10there was some
14:14limitation there as well in in the sense
14:16that uh
14:18uh things uh still existed in terms of
14:21uh uh
14:25the exact terminology would be uh
14:28you know there wasn't really a lot of
14:30distributed uh
14:32uh mechanism available uh so
14:36uh things had to be sent uh to some uh
14:39remote location if you wanted to offload
14:42or
14:42send some computation somewhere then you
14:45have to send it to a remote location
14:48run it and then get the results
14:51but there came this next phase where
14:55things got a little bit more distributed
14:57so the grid computing
14:59uh faces uh such a
15:02architecture where where
15:06the the computing itself became quite
15:10distributed
15:11so you had
15:14different pcs in different networks
15:19and you could actually distribute a
15:21particular computation
15:23workload among these different machines
15:26and that came to be known as the grid
15:29computing phase
15:32and finally
15:35of course grid computing has its
15:37advantages
15:38and that also makes it quite complex
15:42to manage the arc manage the resources
15:46in the architecture because now you have
15:48to
15:49decide how to allocate resources in this
15:52large
15:53distributed architecture right so that
15:55that was
15:56uh one of the challenges there uh
15:59then came uh cloud computing uh and it's
16:02it's a client server architecture
16:04it's a centralized architecture right so
16:06you have uh
16:08you have a large pool of resources that
16:10is sent located in a
16:12uh in a particular place and uh you as a
16:16user
16:17uh using any any of these machines
16:20like pc or mobile device
16:24uh could access uh a set of resources
16:27from this
16:28cloud but
16:32but it's but you have to remember in
16:33terms of if you compare it with grid
16:36computing it's
16:36it's still a centralized kind of
16:38architecture and not a distributed
16:40architecture
16:41so that's the key difference between
16:44grid computing and cloud computing
16:48and it is a little bit different uh you
16:50may think that it is uh
16:52it kind of is similar to mainframe
16:56computing
16:57uh but uh one of the key differences is
17:00that
17:00um there is uh
17:03the the amount of resources that are
17:05available in in cloud computing are much
17:07larger in
17:08comparison to the mainframe computing
17:11phase
17:12uh and of course there is uh uh
17:15there is this enabler uh virtualization
17:18which is also
17:20a part of uh cloud computing which makes
17:23it
17:23uh quite seamless in you know sharing
17:26the resources
17:28yeah so that's that gives a uh
17:32wide picture on uh how the evolution of
17:36cloud computing came came by
17:41uh now let's look at uh the architecture
17:45of
17:45cloud computing so as we know
17:49there as we have already you know heard
17:52that there is there are lots of services
17:55that are provided
17:56to users
17:59and and these services can be of any
18:03type
18:03yeah so based on what kind of services
18:07you are
18:07requiring uh the cloud computing
18:12is is is
18:15it has a layered architecture uh and uh
18:19you can find different kinds of
18:20uh uh you know different kinds of
18:24services provided by different
18:27layers in this architecture so if you
18:29look at it there are
18:32there are primarily three layers to it
18:35but you can also have some extra layers
18:39that we look at so firstly i would uh
18:43talk about this layer called application
18:46uh
18:46which is also called as uh software as a
18:49service layer
18:50so if uh if somebody hasn't heard of
18:53this
18:54already it's it's quite popular
18:57and anybody who has heard of cloud
19:00computing should
19:01also have heard of this but
19:05typically what software as a service
19:08allows is you can actually
19:11so what what has cloud computing now
19:14uh done uh for uh user is that
19:18instead of uh having
19:21all those resources or applications
19:23running on their own
19:26on their own device they can now run it
19:29on on a
19:31cloud right a set of nodes that are
19:33remotely
19:34located um so uh
19:38so basically you can have applications
19:40that can be
19:41run remotely or you can have a set of uh
19:45hardware resources that you can access
19:48from a remote
19:48uh node uh so that that's where
19:52the uh uh sas software as a service
19:55comes into picture
19:56where uh the users can now run the
19:58applications remotely yeah so
20:00so this part of the uh architecture of a
20:04cloud only
20:07provides to the user the application
20:11so if you want to run some application
20:14this is what is a service that you will
20:17request from the cloud and the cloud
20:20would uh
20:22there are there are multiple ways again
20:24uh
20:26the software as a service can also be uh
20:29provided so we will just briefly look at
20:31that as well
20:32uh in a later slide so then
20:36um that's that's only the application
20:38part right so now
20:40uh you can also access more services
20:43right so one of the
20:45uh other things that you can access is
20:48called
20:48infrastructure and in infrastructure as
20:51a service is another
20:53layer of this architecture uh
20:56so what is done here is you
20:59are not only uh here you are not asking
21:03for the application
21:05uh so to say but you are asking for
21:08the computing resources so you are you
21:11are now asking for
21:12uh these set of resources that are
21:15virtualized
21:16on a set of nodes that are available in
21:19the data center
21:21uh and you you will get a logical uh
21:24set of resources uh uh from the uh
21:28that are derived out of the physical uh
21:32compute nodes that are available uh in
21:34the data center
21:35so um uh this this is
21:38uh this is what comes as part of the
21:40infrastructure
21:42uh it's the compute resources so it it
21:44on not
21:45does not only have compute nodes that is
21:49uh you know processing uh resources but
21:52it also has storage
21:53uh resources it also has
21:57some network resources so so all of
22:00these
22:00combined together uh can form the
22:03infrastructure as a resource as a
22:05service
22:06uh that is provided by the uh cloud
22:09provider
22:10uh so of course when when these uh
22:13things are provided one of the things
22:15that needs to be
22:18ensured is that there is a guaranteed
22:20level of processing
22:22that is provided by the uh
22:26cloud provider and uh these get
22:29the it's it's not just about processing
22:31but you also need
22:32some bandwidth reservations that need to
22:35be made
22:36uh so that the amount of data exchanges
22:40that happen
22:41uh can can be done uh you know without
22:45too many delays
22:46uh so uh there is a certain level of
22:50service that also needs to be provided
22:52and that's part of the service
22:55agreements that are made between the
22:57cloud provider and the
22:59user so moving on
23:02um we can add a little bit more to this
23:05infrastructure as a service
23:08uh which is uh you know adding uh things
23:11like
23:12uh an operating system uh
23:15and so so you add an additional stack
23:18over this
23:19uh that's that's the os and also there
23:22are some
23:24other set of service applications that
23:27you could add as
23:28services so basically you build
23:32a software stack on top of the hardware
23:36uh resources that are available as
23:38infrastructure as a service
23:40and that forms uh kind of service that
23:43is called as platform as a service
23:45so that's you know going one step
23:48further so now you can
23:50have the entire uh stack from
23:53hardware to the os to
23:56some set of applications that you want
23:59to
24:00access uh run on your cloud
24:03so that the the this this kind of uh
24:06service can also be accessed from the
24:09cloud
24:11uh and uh finally you uh there is also
24:14this
24:14uh new this there is this
24:18additional service that also uh is
24:21available which is called uh
24:23data storage as a service yeah so you it
24:26it means that
24:27now we are not thinking about uh
24:29computing
24:30but we are more bothered about
24:34the storage aspect so a consumer may
24:37need only
24:38uh some amount of storage
24:41uh from the cloud and that's that's the
24:44service that it requires
24:45so that's the service that's called as
24:49data storage as a service and
24:51it of course does not only include the
24:54storage part
24:55but also it needs some bandwidth
24:58associated with it
24:59so uh so that you know the storage
25:03uh the whole process of storage and
25:05retrieval happens
25:06uh uh in in good time so that
25:10uh uh kind of uh you know
25:13summarizes how uh a cloud architecture
25:16looks like and what are the kinds of uh
25:19services that you uh can avail from the
25:24cloud so this is that it's because there
25:27are these different layers that's why
25:28it's called a
25:29layered architecture yeah so any any
25:32questions until now
25:41okay
25:44yeah please uh if uh yeah
25:47yeah you could unmute and ask or
25:50put your questions in the chat anything
25:52is fine
25:58okay ah so yeah i i'll i'll uh you know
26:01periodically kind of look at the chat
26:03and see if there are some
26:05questions otherwise you could also ask
26:09uh okay moving forward then um
26:14so uh just to give you how uh these
26:17layers uh look like uh
26:20let's let's kind of start with the
26:22easier one which is the
26:25application software as a service
26:28so the architecture now looks like
26:33this so on premises or on
26:37in in some office setup now
26:40you would have these users uh which
26:43would be use uh
26:44the users would be using some browser
26:48uh as a client and uh
26:51they they would be accessing the
26:53application which is actually
26:54running on the cloud yeah so this this
26:57is one kind of
26:57uh software as a service that uh
27:00we see uh quite a lot um
27:04so this this is a very typical software
27:07as a service
27:08uh uh that that is encountered
27:12now there is another kind of uh uh
27:15software as a service that is also
27:17possible
27:18and here is a situation where
27:23the user is actually running uh some
27:25application
27:27on on on the device that
27:30he or she may be using and
27:34uh that application may uh seek
27:37some extra functions uh or extra
27:41uh some set of uh you know
27:45application stubs or whatever
27:49that is not there in the uh application
27:53uh the version or the the part of the
27:56application that the user is
27:58executing locally may be accessed from
28:02the cloud yeah so that's that's also one
28:04kind of
28:05software as a service uh and here
28:09um so one of the examples here is
28:13you know your your apple itunes
28:16uh let's say so uh there
28:19uh you you have a lot of uh
28:22uh a lot of audio
28:26um a lot of media files that you have
28:29and there may be some uh
28:33media file that you don't have on your
28:37application and but you want uh
28:41that that's something that you want to
28:43access and
28:44that's that's part of the cloud so there
28:46there may be some
28:48uh some other functionalities that are
28:51available
28:52uh also that could be accessed from the
28:55cloud that is not part of the
28:58application already so so there are
29:00there are
29:01some uh scenarios like this as well
29:04where
29:04you can have the uh software as a
29:08service kind of
29:09setup now uh the platform as a service
29:13uh as i uh already mentioned
29:17uh in the previous slide it's it's kind
29:19it's as we
29:21uh remember it it's this uh
29:23infrastructure as a service plus the os
29:26and some required services right so so
29:29that is
29:30what this looks like so you you would
29:33have
29:34a lot of the
29:38infrastructure which is the scalable
29:40infrastructure
29:42including the compute nodes storage and
29:46network resources uh and you may also
29:50have a
29:51other set of applications uh including
29:55uh the os okay so so there could be
29:58a entire uh integrated development
30:02environment also that's available
30:04where where things can be uh mod
30:08where new applications can be developed
30:11so they
30:12it the platform as a service will also
30:15provide you
30:16other uh functionalities like data
30:19security
30:20uh because security is one major concern
30:24in
30:25cloud services so data security is some
30:28as
30:28is an important uh uh functionality that
30:32needs to be provided there can be uh
30:36mechanisms that are in place for backup
30:39and recovery
30:40yeah so so assuming that uh something
30:43some some execution was happening um
30:47and suddenly the node
30:51the node on which it was happening fails
30:53so there needs to be some mechanism to
30:56recover the computation
30:59and so that that's where you know backup
31:01and recovery come into place
31:03and there can be various kinds of
31:05applications that are hosted
31:07so so you you can uh
31:10you can have this entire stack
31:14as part of the platform as a service
31:17so this is this is how uh
31:20internally uh these services would look
31:23like
31:27okay so with that understanding uh so we
31:30have
31:30we have gone through uh what
31:34is cloud computing uh we have seen
31:37uh what are the different
31:42architectures in cloud computing i mean
31:45the different layers in cloud computing
31:48and
31:49we have also seen in terms of evolution
31:53how
31:53cloud computing came into existence now
31:56uh we we have to uh also look at
32:00uh the different types of cloud
32:03computing
32:04uh and and here uh it is more
32:08uh you know the
32:12the the different types are uh in terms
32:15of
32:15how uh they are accessed yeah so how how
32:19they are
32:20uh available to the user uh so
32:24the three types of cloud computing that
32:26are
32:28very that are used in various scenarios
32:31are
32:31a public cloud a private cloud
32:35and and there's something called a
32:37hybrid cloud
32:39so uh public cloud is uh nothing but
32:43uh you know the the scenario where
32:46you're actually
32:47there are these multiple users which are
32:49actually uh asking a third party
32:52uh cloud service provider
32:55some resources or some services like we
32:58already saw
32:59yeah so that is a public cloud whereby
33:02everything every service or
33:06resource that you are trying to access
33:08is
33:09remote so it's it's external to the
33:12cloud
33:14it's it's an external cloud so it's it's
33:16not something that is
33:19part of the
33:22setup where the user is user is located
33:26so that's that's the uh
33:29public cloud so the computing resources
33:32once the user
33:33asks for some service
33:36in the public cloud uh this third party
33:40provider which is an off-site uh
33:43third-party provider would allocate some
33:45resources and
33:47provide whatever
33:50service the user wants
33:54and uh one of the things that could
33:57happen there
33:58is which is uh quite natural that there
34:01are multiple users now
34:02trying to access some service from
34:07uh the cloud uh so there could be
34:11uh you know dif applications from
34:13different users
34:14that are mixed together in the cloud
34:17right so
34:18um if you remember that
34:21there is they are given virtualized set
34:24of resources
34:25and there can be sharing of resources
34:28that are
34:29seen right so when there is sharing of
34:32resources
34:34there can be applications that are
34:37mixed together and executed within
34:41the cloud so you can have
34:46a couple of applications that are
34:48running on the same
34:50uh set of um
34:54they could be sharing uh some physical
34:56resources
34:57while executing yeah but of course in a
35:00virtualized fashion so
35:02that's something that the public cloud
35:04would uh
35:06have uh also
35:09uh you know going a little
35:14in the other direction from public cloud
35:16uh you can see something called the
35:18private cloud which is
35:19internal uh to an organization so
35:22basically this is
35:23this is a type of
35:27cloud computing which is built
35:29exclusively for
35:31a client yeah so there is one
35:35organization which wants uh cloud
35:40technology to have uh
35:43you know store some of its very uh
35:46private uh
35:47information uh that private information
35:50of the
35:51organization and uh in in that situation
35:55uh it may not be feasible to send the
35:58data
35:59or do anything do any computation by
36:02accessing services from
36:04a public cloud and that's where the
36:08private cloud comes into picture so the
36:10advantage of having
36:12the private cloud is uh you know there
36:14is full control of the data
36:15yeah because now uh the
36:19data remains within the purview of
36:22the organization uh so that
36:26there is absolutely no possibility of
36:28data being
36:29uh tampered with uh by
36:33somebody some unknown entity
36:36um although cloud providers are
36:40supposed to also take care of these
36:42things but still there can be breaches
36:46so security of course is also something
36:50that
36:50you know can be now taken care of
36:54uh by the uh by the
36:57uh user or the cloud the organization
37:00itself
37:01and um uh there is uh
37:04as opposed to what would happen in a
37:06public cloud where
37:08if you are asking for a service there is
37:10a agreement that is made between the
37:13user and the cloud
37:17provider based on how much you pay there
37:20is a certain level of quality of service
37:22you get but in the case of an private
37:25cloud that is
37:26not too much of a concern because you
37:28get full quality of service right you
37:30are not
37:31there is no such uh
37:36mechanism of uh giving quality of
37:39service based on
37:40how much person pays right because that
37:43it's completely owned by the
37:45uh organization itself
37:48uh so uh that that's that's the
37:50advantage that you have but
37:52of course that uh the overhead that
37:54comes into
37:55uh uh this this kind of private cloud
37:59setup is that
38:00the or the company or the organization
38:03itself has to now uh manage and build
38:07this
38:08cloud right so it's it's uh
38:11as i mentioned in the in an earlier
38:13slide
38:16the the owners of uh setting up this
38:19infrastructure is now
38:21on the organization and
38:24uh and there is a lot of overhead in
38:28not just setting it up but managing it
38:31over a
38:31period of time so so that's that's the
38:34burden that comes with it but of course
38:37you get uh some of the other advantages
38:40that
38:41you may have to forego in in the public
38:44cloud
38:45scenario uh and and finally
38:48there are you can have a uh you can have
38:51best of both worlds kind of scenario
38:54uh if you are using something called as
38:56the hybrid cloud
38:58uh whereby you can have certain things
39:02um uh you know put into uh a private
39:05cloud
39:07that you want to maintain and some of
39:10the aspects you can push to a public
39:12cloud
39:13yeah so so there may be some some
39:16information or some kind of applications
39:18that you want to run which are
39:20not very critical uh and and it's okay
39:23if there are some uh you know if some
39:27external entity uh somehow
39:31has access to it but uh there
39:34there are some uh uh
39:37some uh uh info there is some
39:40information that uh
39:41or that an organization may not want to
39:44reveal in any manner right so that those
39:48kinds of
39:49scenarios you may want to use something
39:52called the hybrid cloud model and that
39:56that gives you uh both of these things
39:58so you you will you will have full
40:00control
40:01on things that you uh deem to be
40:04important and
40:06uh you you will let go of the control
40:08and also some
40:10burden of maintaining a very large
40:13cloud infrastructure by sending some of
40:16the unimportant data to the
40:18public law yeah so
40:22that of course gives you some
40:23flexibility uh but
40:26one of the uh interesting things that
40:29happens here as in the case of
40:32distributed
40:33setup is that now
40:36there is this challenge of uh
40:39determining how to distribute your
40:42applications
40:43yeah so um you have to
40:48find out what are those specific
40:52applications or parts of applications
40:55or types of data that you want to
40:59have as private and or you want it on a
41:03private cloud and
41:04the ones that you want on the public
41:06cloud now how do you distribute this
41:08is is a very um
41:12interesting question i mean there could
41:14be several techniques to
41:16do this some could be you know very uh
41:20static uh mechanisms of doing it
41:23like you decide up front that this is
41:26how
41:26the distribution uh that's going to be
41:29and and you uh distributed it
41:32accordingly
41:33but there could be a very dynamic ways
41:35of doing this as well so therefore there
41:37are very interesting uh
41:39uh challenges there uh of how to
41:42distribute this
41:43uh application set of applications among
41:46the various
41:49kinds of types of clubs
41:55so let's look at
41:58now let's look at something some
42:02features of cloud computing i mean we
42:03have looked uh
42:05briefly touched upon them uh in in some
42:08of the earlier
42:09slides but let's focus on them now
42:12because
42:15some of these aspects are uh quite
42:17important
42:18um in they are
42:22they seem to be advantages in terms of
42:25uh in terms of a
42:28scenario where your track where a user
42:32via his uh desktop or laptop is trying
42:36to access
42:37some service uh in in that scenario
42:40there are some features that the cloud
42:42computing provides
42:44um and and once we look at it then we
42:47will
42:48go on to see what may be some of the
42:51issues with cloud computing and and
42:53that's where the
42:56motivation for fog and edge computing
42:59comes into
43:00picture so um as we have seen
43:04uh there is a scalability uh
43:07in in cloud computing and
43:10there is also something uh which is
43:13known as on-demand uh
43:15uh you know provision of services so
43:18services uh uh if if there are some
43:21services requested from the
43:24cloud uh there is an on-demand way of
43:29providing the resources uh and executing
43:32those services
43:34on the cloud yeah so this it's an
43:35on-demand mechanism
43:37and uh uh of course
43:40scalability uh means that
43:44based on your requirement at a
43:47particular point in time
43:49uh and and the uh kind of agreement that
43:53you have with a
43:54cloud provider your resources could be
43:56scaled
43:58and assuming that you were using
44:01uh in terms of uh virtual machines
44:04terminology let's say you were using a
44:08couple of
44:09vms to execute some very large
44:13workload and that workload
44:16became
44:19was scaled to some high performance
44:23kind of scenario and then these two vms
44:26are not
44:27enough you you you require more than two
44:30vms in that scenario
44:32uh it would be possible to scale the
44:35resources
44:37also you could scale it over several
44:40data centers
44:41so there there is a lot of scalable
44:44uh scalability option that's available
44:47in terms of
44:48cloud computing um
44:52so uh the other aspect that's also
44:56uh you know true about cloud computing
44:58is that it is uh
44:59it provides this user centric interface
45:03so uh the user can now
45:07access uh you know these services
45:11uh either by internet browsers or web
45:15services
45:16that are available so it's it's uh
45:19it's more uh in a user-centric approach
45:23of uh
45:24providing these services so uh that's
45:27that's one
45:28feature that cloud computing has
45:33as we have already talked about quality
45:36of service
45:36right so one of the things that
45:40cloud computing strives to
45:44achieve is every user uh based on the
45:48agreement that he has there is a certain
45:51level of quality of service that is
45:55demanded and the cloud provider has to
45:59guarantee the
46:01level of quality of service that the
46:03user has
46:04asked for so guaranteeing
46:09quality of service means that
46:12let's say let's say let's say we take an
46:14application and
46:15uh um and executing
46:19an application uh the the
46:24contract is that uh uh
46:27every uh gb of data
46:31needs to be processed within some x
46:33amount of time
46:34okay so that if that is the scenario
46:37then
46:37uh that qos is
46:41is guaranteed in terms of whatever
46:44resources are provided by the cloud
46:46provider so so it is uh the owners of
46:50the cloud
46:50cloud provider now to uh
46:54choose what is the what are the uh
46:57resources required and how how much
46:59resources are required
47:01to uh guarantee these uh set of
47:04uh these uh service level agreements
47:07that are
47:07uh that that need to be ensured
47:11for the user yeah so it could be in
47:13terms of uh
47:14how much cpu performance you want uh how
47:17many sets of
47:18cores you want um the bandwidth
47:22that needs to be there for uh you know
47:25exchange of data
47:26what is the memory capacity that is
47:29required
47:30to run these applications or
47:34access uh access the platforms so
47:38all of these things uh are part of the
47:41cloud computing framework that uh and a
47:44provider has to ensure this and and this
47:48is where the
47:49uh the part of pricing comes into
47:53picture so
47:54um as we know uh nothing is free
47:58uh there is there is some uh cost that
48:02is incurred
48:03uh so you're not you're saving on the
48:06very huge cost of
48:08not setting up the cloud infrastructure
48:11and you're seeking that service from a
48:14third party
48:15so there needs to be uh some
48:20benefit for the cloud provider as well
48:23and that's where
48:24pricing mechanism comes into play and so
48:28every cloud provider has a price pricing
48:30mechanism
48:32uh be it uh in terms of the
48:35data uh you know that you transfer
48:39between
48:39uh the uh between your
48:43device and the cloud node
48:46uh that you use the amount of or
48:49the capacity of the node that you uh
48:52want
48:52um so all of these things
48:57you know play an important role in
48:59deciding what is the price that you pay
49:01for accessing these services
49:03so there there are uh if you look at uh
49:05you know amazon
49:07um google cloud google cloud and various
49:11of various
49:12uh such services uh you can see that
49:15there are uh
49:16they price uh uh their services
49:19and they have they have a uh fixed uh
49:22pricing
49:23there are there are various kinds of
49:25pricing mechanisms that are set
49:27um whereby users can access different
49:29kinds of services so there may be some
49:31services which have
49:32uh free access yeah even even if you see
49:36on some of these well-known
49:40cloud platforms some some services may
49:42be absolutely free so you don't have to
49:44pay anything
49:45but there there can be some other
49:48services
49:50or there may be services where up to a
49:53certain
49:54quantum of data it is free and then
49:56after that it becomes
49:57chargeable and so there can be various
49:59such pricing mechanisms that are
50:01there and that's part of the
50:04cloud computing setup
50:09so in terms of pricing there are three
50:12dimensions uh as i uh already mentioned
50:15that are
50:16uh taken into consideration uh
50:19storage is one aspect so uh you have to
50:24pay uh for the storage
50:27uh the amount of storage resources you
50:30access uh based on uh
50:34the amount of data that you want to
50:37store
50:37and it's g it typically is in
50:40gigabytes and over a monthly period and
50:43so that's that's one
50:46dimension of the pricing uh there are
50:50bandwidth resources that you want so
50:53it's not just
50:54you know storage or compute but you also
50:57need to
50:57transfer data in and out of the platform
51:00service that you are
51:01trying to access from the cloud so um
51:04if if you have to do that in the
51:09in the strict amount of time that you
51:12are willing to spend for doing it then
51:15bandwidth is an important criteria that
51:17needs to be satisfied
51:20and then there is compute uh uh
51:24pricing so uh it it could be based on
51:28you know how many
51:29uh course you want to access for how
51:32much
51:32time uh so that could be one way of
51:37determining how what the compute price
51:39is so as as
51:41as is mentioned here the uh you could be
51:44charged based on the time units that you
51:49need to run an instance
51:52or or it could be you know running an
51:56entire application or you could also
51:59maybe
52:00you know based on whether you are
52:03accessing an entire
52:05set of resources for for for servicing
52:08your request
52:09so um so
52:12compute uh basically uh
52:16involves you know how much time you need
52:20to
52:20access those resources so it's it's an
52:23actually
52:23it's a very interesting uh scenario and
52:26there are
52:26lot of research
52:30works also that have been done in this
52:32area where
52:33people have tried to optimize how
52:38how we can allocate resources
52:42to multiple users such that
52:47you know the pricing can
52:50the the total price uh can be minimized
52:53yeah so so that the users are not
52:57charged too much so there can be there
52:59can be
53:00the the allocation also plays an
53:02important uh role in
53:04deciding uh the pricing yeah so the
53:08there are some interesting trade-offs
53:10that can be
53:11uh looked at there in how how we can
53:15allocate these resources so that these
53:17prices can be
53:18minimized yeah so uh
53:22moving on to uh the challenges
53:26in cloud computing
53:30so one of the very
53:35very negative aspects that you know
53:38people do
53:39ask about in regarding cloud computing
53:43which people you know think about an uh
53:47as an aspect which is against cloud
53:49computing is this uh
53:51performance aspect so
53:54uh in the performance aspect the issue
53:57is that
53:59the users typically could be
54:04you know located very far away from
54:06these cloud providers
54:08which result in uh excessively
54:11uh long latencies yeah so the
54:14uh if i i don't know if you have uh
54:19in your previous session on iot
54:22applications one of the things that you
54:24may have
54:25seen different iot use cases or
54:29applications
54:30one of the things that you may have
54:31looked at is
54:34the the various kinds of applications
54:37that are there
54:38the the nature of these applications
54:41they may require
54:42different kinds of
54:45timing requirements yeah so um if if
54:49some applications need
54:52to execute in a very short time
54:55uh then having we will we will see some
54:58examples in that area so if
55:02if you require that kind of a
55:04performance
55:05then cloud computing has an issue
55:08because
55:10there is there is a potential for
55:14a long round trip time that that's there
55:18okay so that's one of the things
55:23that needs to be as challenges in cloud
55:26computing
55:27and the other issue that also exists is
55:32the issue of security and privacy which
55:34is very important
55:36in the current
55:40ages because there are attacks happening
55:45various mechanisms are utilized to
55:49attack various
55:52resources various parts of the
55:55infrastructure
55:58also thereby gaining access to some very
56:00important
56:02private information and causing a lot of
56:06you know causing a lot of problems to
56:10cost and
56:12you know issues that come up when there
56:15is a downtime of the
56:17whole crowd cloud resource
56:21so security is an issue
56:24security and privacy both are very
56:26important issues and
56:29the the issue uh arises uh because there
56:32are
56:33there are some concerns in terms of
56:36vulnerability
56:37to attacks especially when the
56:41information
56:42uh is outside uh the firewall yeah so if
56:47if there are not proper
56:51steps followed by the cloud provider
56:55and proper security mechanisms are not
56:59there then
57:00your information has gone out into the
57:06into the network remote network and
57:10and then you don't have any control over
57:12it right so if if the
57:14cloud provider has not put in uh place
57:17the
57:17required mechanisms
57:21to ensure these aspects then then it
57:24becomes
57:25quite vulnerable to attacks especially
57:28also because you know
57:29uh one of the as as we already saw one
57:31of the enabling technologies
57:33for cloud computing is virtualization
57:36and there are
57:36there are these applications uh
57:40running on vms virtual machines and
57:44there could be attack surfaces or attack
57:47possibilities that are
57:49there from one vm to the other
57:52uh which uh which typically
57:56uh there are uh the the chances of that
57:59happening is very low but
58:01there can be still uh lot of um
58:04uh as we as in the security
58:08uh you know field it's
58:11called a side channel so there could be
58:13some side channels that are exploited
58:17uh which can enable these
58:20attacks possible uh and and uh
58:24that that could be one major issue that
58:27you
58:28uh do not want to happen with your very
58:30critical or important data
58:33so that is one thing uh the other aspect
58:36is
58:38control right so
58:43in especially in public clouds
58:46uh you you don't have any control over
58:50the the way things are done so you don't
58:54have any control over
58:55how your data is handled how
59:00the set of applications that you want to
59:04execute are executing so
59:07the control aspect is something that
59:10[Music]
59:12is is absolutely not in your hands
59:15in terms of cloud computing um
59:18unless you are you know willing to do
59:22something like a private cloud or or
59:25in in some ways you know the hybrid
59:27cloud
59:29so other other very important apart from
59:32performance and security performance
59:34security and privacy
59:36the other very important part which is
59:39quite detrimental to the whole
59:42set of advantages that cloud computing
59:45provides is the bandwidth cost
59:47so now assume that there are these
59:51different i mean millions of users
59:55they want to
59:58use the services uh provided by a
1:00:02cloud provider and they are sending data
1:00:06um they are sending their set of data to
1:00:09the
1:00:09cloud for for the cloud to
1:00:13work on it now uh that incurs a lot of
1:00:17cost yeah so every
1:00:18every user has to uh
1:00:22pay for the uh data that he or she sends
1:00:28and also the the amount of data that is
1:00:31handled at the
1:00:33cloud side and the amount of data that
1:00:37goes into the
1:00:38network all of it adds to the bandwidth
1:00:41costs so
1:00:43bandwidth cost is a very important
1:00:47phenomenon which which really
1:00:51has made the cloud computing
1:00:56quite challenging in in the current era
1:00:59of
1:00:59iot so we will kind of see the emphasis
1:01:03of this
1:01:03uh as we go along to
1:01:06the topic of edge computing
1:01:11so reliability is also one aspect
1:01:13although you know
1:01:14[Music]
1:01:17they try to make it very reliable by
1:01:22having uh you know redundancies
1:01:25uh in place uh having uh executions
1:01:29having the executions uh hosted on
1:01:32multiple notes
1:01:33um so that if one fails the other
1:01:37is still there there is still a result
1:01:39from the other but
1:01:40uh it it still uh may not
1:01:44uh be you know available uh 24 cross
1:01:47seven and we have seen
1:01:49uh in a lot of instances like in
1:01:54especially it happens in the u.s where
1:01:57there are certain
1:02:00days where there is there is some kind
1:02:02of off
1:02:03sale offer that's going on due to some
1:02:06festive season and
1:02:08suddenly the entire thing uh
1:02:11the the whole service stops
1:02:15because of the uh large
1:02:18amount of traffic that's going into the
1:02:21uh
1:02:22you know the system so uh so the
1:02:24reliability is definitely an issue
1:02:27so you cannot always uh ensure that
1:02:31cloud computing uh will be uh provide
1:02:34the the services of cloud computing will
1:02:36be available
1:02:38uh all the time of the year
1:02:41so these are these are some of the
1:02:42challenges that
1:02:44cloud computing has
1:02:50so now let's get into some motivations
1:02:52on uh
1:02:55having this background on cloud
1:02:57computing
1:02:58let's see um how uh
1:03:02some of these aspects uh uh
1:03:05do not really sit well when you're going
1:03:08into this
1:03:09iot era yeah so
1:03:13the emergence of iot is one of
1:03:16the important events that is
1:03:20that has happened in the last decade and
1:03:24and things are kind of moving forward in
1:03:28in
1:03:29quick pace so you can see that in
1:03:332003 this is
1:03:38data that is collected from cisco uh
1:03:41document so uh in in 2003 there
1:03:46uh was this uh uh the the number of
1:03:50connected devices were about 500 million
1:03:53and the number of the world population
1:03:55was about 6.3 billion
1:03:57yeah so so here the world population was
1:04:02some orders of magnitude higher than
1:04:06the connected devices now at some point
1:04:09in time here's
1:04:10around 2005 2006
1:04:14uh it the the scales
1:04:18you know tipped and the number of
1:04:21connected devices became more than the
1:04:23number of
1:04:24people in the world so that's that's now
1:04:28uh that's now very significant yeah so
1:04:32uh coming to uh 2010
1:04:35uh you can see it's kind of become
1:04:37double
1:04:38uh so in in about four years from
1:04:41when uh the number of connected devices
1:04:44uh became uh just uh um
1:04:47a little bit more than the number of
1:04:49people
1:04:50uh to the time uh to uh 2010 when
1:04:56it you know that's just a span of four
1:04:59years and it became
1:05:00twice the number of uh almost twice the
1:05:03number of
1:05:04uh the number of people in the world
1:05:06yeah and then this
1:05:08gap is kind of increasing pretty fast
1:05:11so so now it's about
1:05:14in 2015 it was about three and a half
1:05:17times
1:05:18and in 2020 uh the numbers uh
1:05:22seem to be six and a half times around
1:05:25uh
1:05:26very close to seven times uh the number
1:05:28of people
1:05:29so what does this mean so this means
1:05:31that uh
1:05:33now it's not just the user who is trying
1:05:35to connect
1:05:37to the cloud there are
1:05:40and and if you under i mean
1:05:44from your previous lectures you would
1:05:45have understood this concept of iot
1:05:47right so iot
1:05:49means anything can now be connected to
1:05:51the
1:05:52uh to the network uh so
1:05:55uh so if if there are if
1:05:58if practically anything can be connected
1:06:01to the network
1:06:02then uh and if you are seeing these
1:06:05connected uh things or
1:06:08devices uh uh increasing in this
1:06:13in this progression then you can
1:06:17you see that it's it's really going to
1:06:19go very
1:06:20uh upward and the number of users is not
1:06:24the
1:06:25uh the determining criteria now it's the
1:06:29number of connected devices yeah so
1:06:32um if there are uh like 50 billion
1:06:36uh devices plus the users that are
1:06:39connecting
1:06:40um of course i mean this is this is the
1:06:43worst case scenario
1:06:44that all of them connect at the same
1:06:47time but
1:06:49you can very well imagine this kind of
1:06:51scenario happening sometime later
1:06:53so when when these many number of
1:06:56entities are trying to connect and send
1:06:58their data
1:06:59to the cloud that's not what cloud was
1:07:02designed for
1:07:04yeah so uh it was not uh the the
1:07:07very inception of cloud was not uh
1:07:11did not happen keeping in mind the iot
1:07:14scenario
1:07:15so iot scenario happened later and
1:07:19cloud is trying to catch up to that
1:07:22scenario
1:07:23and there are various things that are
1:07:24being done
1:07:26so so therefore there is a challenge
1:07:29that's
1:07:30uh uh that's there because
1:07:33of the uh because of the
1:07:36sheer magnitude of the number of
1:07:40types and number of devices and the
1:07:42different kinds of devices that are
1:07:43trying to
1:07:45connect to the cloud
1:07:48so uh that you know does just does not
1:07:51stop right i mean internet
1:07:54of things and then there is internet of
1:07:57everything
1:07:57right you could have
1:08:02people you could have non-living things
1:08:06everything is now connected and uh
1:08:11everything is trying to send some
1:08:13message across yeah so
1:08:15uh this the emergence of this uh
1:08:19phenomenon of internet of things has
1:08:22really
1:08:23made it quite challenging for
1:08:26the way the systems were designed for
1:08:30cloud computing the way
1:08:34resources have to be managed on the
1:08:36cloud uh
1:08:38and uh you know the way things uh
1:08:41have to be processed so the the way data
1:08:44has to be handled
1:08:46uh needs to be uh taken care of uh
1:08:49uh with with a lot more
1:08:53um you know
1:08:57newer technologies than just what cloud
1:09:00provides yeah so that's that's one of
1:09:02the
1:09:02uh challenges that we have
1:09:06uh so uh i will
1:09:09look at uh the uh look at some of these
1:09:12challenges
1:09:13and uh maybe i'll uh stop
1:09:17after going through a couple of slides
1:09:18i'll stop for some time to
1:09:22see if there are some questions and
1:09:24after that i think we can
1:09:26take a break and then come back and look
1:09:29at
1:09:30the edge scenario
1:09:33so as as now you can see the cloud
1:09:37is facing some of these challenges
1:09:41that we kind of hinted upon
1:09:46in our previous slides so there is
1:09:49uh the volume challenge yeah there is a
1:09:52lot of data that's coming
1:09:53because there are a lot of entities that
1:09:55are connected now
1:09:57um one of the
1:09:59[Music]
1:10:01one of the data points that's there is
1:10:03that
1:10:05when cars start getting uh you know
1:10:09connected i mean of course in lots of
1:10:10places cars are already
1:10:12connected fully uh and they are they are
1:10:15also
1:10:16uh in some places they are also uh
1:10:20almost like autonomous but uh uh
1:10:23if there if there is this fully
1:10:25autonomous car that's uh there
1:10:28uh and there are a lot of sensors that
1:10:30are placed within
1:10:31uh these cars uh the the amount of data
1:10:35that it generates
1:10:37is is is very huge so it's it's in the
1:10:40order of uh
1:10:41gbs per um
1:10:44per hour or something i don't exactly
1:10:46remember the
1:10:48figure but it's a very high value
1:10:52so if you have to deal with that kind of
1:10:56data
1:10:57that's coming from just one uh
1:11:00one device yeah one
1:11:04device which is the car in itself
1:11:07uh that's that's a lot of data that you
1:11:10are getting and you have to
1:11:12make sense out of the data because these
1:11:15sensors are not there inside the car
1:11:18uh just for the sake of you know
1:11:22putting it there because it looks good
1:11:26and you're getting some data but there
1:11:28needs to be some
1:11:30sense made out of it there needs to be
1:11:32some processing that needs to be done
1:11:34and imagine that all of this processing
1:11:36has to happen
1:11:37uh over the cloud so which means every
1:11:40car needs to send that much data to the
1:11:43cloud
1:11:44um and that's that's going to be a huge
1:11:47challenge yeah it's not
1:11:48it's not feasible to do that so
1:11:51so that's one of the very major
1:11:53challenges that
1:11:55cloud computing has ah of course
1:11:58um as as we already know
1:12:01uh these end nodes or
1:12:05devices are not
1:12:08anywhere proximally near to the
1:12:11cloud or the data center
1:12:15and that could incur high latency
1:12:18and again taking the example of a car
1:12:22there could be some
1:12:25services or some functionality that it
1:12:28wants to execute
1:12:30[Music]
1:12:32i mean typically currently the cars have
1:12:34not still
1:12:36[Music]
1:12:39there are no scenarios in cars still
1:12:41that people have
1:12:42although people have identified
1:12:44scenarios inside the car and we will
1:12:46look at one such example
1:12:48but still people stay away from you know
1:12:51some of the critical functionalities of
1:12:53the car because
1:12:54uh you know car uh is a safety critical
1:12:57system
1:12:58uh and uh you do not want any
1:13:02uh untoward incident happening right so
1:13:05if if
1:13:06if things don't happen at the right time
1:13:08uh things can
1:13:10get really catastrophic in in in the car
1:13:13scenario so
1:13:15high latency is one of the issues
1:13:18uh which a real-time scenario like a car
1:13:23and cannot really tolerate yeah any any
1:13:26such
1:13:26device that needs a real-time
1:13:29[Music]
1:13:31thing that it it cannot be done uh using
1:13:34uh the cloud computing
1:13:35model uh one other example could be like
1:13:38your
1:13:39your monitoring um let's say
1:13:42uh some fire uh breakout in in a
1:13:46in a house and and you want to
1:13:49uh basically send that information
1:13:53and get back some response quickly yeah
1:13:57now uh if uh understanding that
1:14:01there is such a thing and the result
1:14:05has to happen you know the the whole
1:14:08processing of the data
1:14:11and inference of that data has to happen
1:14:14by going to the cloud and then you get
1:14:16coming back
1:14:17with the inference then that could uh
1:14:21cause a lot of delay and of course that
1:14:23have also affects the subsequent
1:14:26help that you get from the uh
1:14:29you know fire uh department so
1:14:33so uh there are some situations where
1:14:35high latency is
1:14:36completely uh not
1:14:39accepted and but there could be some
1:14:43scenarios some applications where
1:14:46it's okay to have some latency like
1:14:50let's say you
1:14:50have some application like
1:14:56monitoring uh the quality of air at some
1:15:00location in the city yeah and and
1:15:04here and you want to make some
1:15:08inference based on that so you're
1:15:10collecting data
1:15:12and you're making some inference of how
1:15:15does the
1:15:17quality of air
1:15:20change based on some activity that's
1:15:23happening in the city
1:15:25so now here it's not really necessary
1:15:27for you to
1:15:29get the inference quickly yeah you don't
1:15:32uh you collect data you process it and
1:15:35you then you come up with
1:15:38some inference that yes that activity is
1:15:41causing
1:15:42the air quality to deteriorate
1:15:46and then you take decisions as to policy
1:15:49decisions as to whether
1:15:50uh some what needs to be done to
1:15:54set it right so that in in such
1:15:56scenarios latency is not an issue so
1:16:00so there so there are
1:16:03scenarios like this which are not really
1:16:07which not require real-time constraints
1:16:09uh where
1:16:11it is okay to suffer some latency
1:16:14uh go over the cloud and then get back
1:16:16your
1:16:17results yeah so so there are multiple uh
1:16:20scenarios based on the scenario uh
1:16:24it could be a challenge to uh use
1:16:27cloud for your application
1:16:32um the other aspect is uh
1:16:35the transportation or communication cost
1:16:38so uh large amount of data
1:16:42going over the um you know
1:16:45network is is a challenge in itself
1:16:49because of the costs involved with it
1:16:51you know transmitting that much amount
1:16:54of data
1:16:56is is not does not come for free so
1:16:59there is this
1:16:59cost involved especially in the case of
1:17:03as i mentioned in the case of the car uh
1:17:05if you want to
1:17:06keep transmitting that much amount of
1:17:08data that's going to
1:17:10uh incur some cost for you so that's
1:17:14that's not something that's
1:17:16advisable and
1:17:20resiliency is also something that's not
1:17:23always practical
1:17:28especially let's say you are you're
1:17:30looking at some scenario where
1:17:32uh there is no way that the
1:17:37remote node which needs to do some
1:17:39operation and send you some result
1:17:41just goes down for some time um
1:17:44so a cloud it can happen
1:17:47you're not going to have uh uh
1:17:51resilient uh i mean although there are
1:17:55uh mechanisms that and
1:17:58techniques that are uh put in place so
1:18:01that the
1:18:02uh resiliency is maximized but
1:18:05uh that does not guarantee that it's
1:18:07going to be 100 resilient so
1:18:10ah so for some scenarios you may not
1:18:13want to take the chance
1:18:14so receive resiliency is also an issue
1:18:16that some scenarios you may not
1:18:19want to affect your
1:18:23execution so these are some of the
1:18:26challenges that cloud computing has
1:18:29uh so uh i will
1:18:32stop with this slide i'll just go
1:18:35through this slide and i'll stop here
1:18:37because this is the motivation to
1:18:41why we need edge computing
1:18:46and the reason being as i already
1:18:49mentioned there are
1:18:51there are a lot of devices that are
1:18:55[Music]
1:18:56currently connected and
1:19:00obviously the emergence of iot has made
1:19:02that possible
1:19:04uh the data so one of the data points
1:19:06that
1:19:08is available says that by 2025
1:19:12uh this these
1:19:15these many devices will generate
1:19:18something like 79.4
1:19:20zeta bytes of data that's that's 10 to
1:19:23the power of 21
1:19:25uh you know so that's that's a lot of
1:19:28data
1:19:30and that actually would require
1:19:34something like thousand data centers yep
1:19:37so
1:19:38um that's that's not a very uh
1:19:42uh you know positive scenario right i
1:19:44mean it's it's
1:19:45uh handling this much this much data is
1:19:49very challenging and
1:19:52over and above that you have the cloud
1:19:55computing
1:19:57cara fee characteristics that we just
1:19:59saw which is
1:20:00a high latency
1:20:03scenario with due to the uh
1:20:06due to the location of where the end
1:20:08device is and where the cloud
1:20:10node is uh there is a centralized model
1:20:13of computing right in the
1:20:16in the cloud scenario so um because of
1:20:19the centralized model of computing
1:20:21um there are uh scenarios where the
1:20:25uh the the cloud node could go
1:20:29uh could could just be non-functional
1:20:33and then you don't have any other way of
1:20:36executing your required services
1:20:40uh one other important aspect that cloud
1:20:43computing
1:20:44lacks is the location awareness yeah so
1:20:47so assuming that
1:20:49you send some uh you you want some
1:20:52service to get
1:20:53uh uh executed for for some purpose
1:20:57uh cloud does not have uh you know
1:21:01automatically there is no location
1:21:03awareness that it has it's it's just
1:21:05getting some data executing it and
1:21:09it does not have information on what is
1:21:12the
1:21:13uh situation in the nearby environment
1:21:17from where that device gen uh is
1:21:19requesting for a service
1:21:21and so that that information is not
1:21:24something that cloud computing can
1:21:27leverage
1:21:28which is possible in in lots of us
1:21:31applications so
1:21:32location awareness is not something
1:21:34that's there
1:21:36the other important aspect is cloud
1:21:39computing
1:21:41absolutely there is no support for
1:21:43mobility
1:21:44uh so so it becomes very challenging for
1:21:50some mobile node in the environment to
1:21:54you know reach out to cloud and try to
1:21:56execute some service
1:21:58and um as as we already discussed it's
1:22:02uh well there are there is vulnerability
1:22:05to security attacks
1:22:07so in short summarizing the the main
1:22:11three aspects that uh
1:22:13kind of motivate uh from you know moving
1:22:16from
1:22:16a cloud computing model to something
1:22:20little bit more distributed
1:22:22is uh are these three the volume
1:22:25of data the latency that the data uh
1:22:28experiences and uh the bandwidth yeah so
1:22:32uh these these three aspects do not uh
1:22:36uh uh you know do not fare well in uh
1:22:39when we are looking at
1:22:40cloud computing kind of model uh so
1:22:43with this i think uh uh i will stop
1:22:46now and uh uh if there are some
1:22:49questions
1:22:50until now we can take that and
1:22:53we can break for lunch and come back
1:22:56later
1:23:01yeah participants feel free to unmute
1:23:03and ask the questions
1:23:06yeah so am i honorable
1:23:10yeah yes yes yes good afternoon thank
1:23:12you so much for a detailed insight it
1:23:14was really clarifying to understand how
1:23:17cloud computing works
1:23:18so uh i'm working in the field of iot
1:23:22security and i
1:23:24i would like to understand that as a
1:23:26person who's doing research in iot
1:23:28security we know that
1:23:30like there are many security issues and
1:23:33we do
1:23:34understand that the resources available
1:23:37in an iot device
1:23:39is like limited so we know that
1:23:42if i want to run some uh high-end
1:23:46computations like machine learning or
1:23:48deep learning or an iot device
1:23:51it's it's it's practically uh not
1:23:53possible considering we don't have that
1:23:55much computation for our
1:23:57and that is when we uh take help from
1:24:00cloud computing but
1:24:02the question is like when we are trying
1:24:06to secure the iot device but we are
1:24:09going
1:24:09to uh towards uh cloud computing
1:24:12which itself is not 100 secure so is it
1:24:16not a trade-off that
1:24:17whether uh the security is increasing in
1:24:20uh
1:24:20uh like is it actually increasing or
1:24:24it's like we are thinking it's
1:24:26increasing but it's adding to
1:24:27more uh vulnerabilities in the system
1:24:32yeah so that's that's a very good
1:24:33question uh so
1:24:35if i summarize your question uh
1:24:38you're you're asking whether um
1:24:42i moving towards an iot
1:24:45uh frame iot based uh
1:24:49scenario um and and maybe a more
1:24:52distributed scenario
1:24:55are we doing any better in terms of
1:24:58security
1:24:59uh because the cloud model itself was
1:25:02not uh
1:25:04entirely secure so are we doing better
1:25:06in terms of security yeah that's that's
1:25:08that was your question right
1:25:09yes sir yes so yes of course so
1:25:13um there are limitations uh
1:25:16so there are
1:25:20a lot of issues
1:25:23that are added uh because
1:25:26now these devices are uh
1:25:29[Music]
1:25:31these these different devices are points
1:25:34of
1:25:34attacks and uh that that
1:25:37increases the attack surfaces
1:25:42uh but one of the uh important aspects
1:25:46that is and of course there are a lot of
1:25:47other aspects that you mentioned like
1:25:49you know executing machine learning
1:25:52algorithms which in itself is a very
1:25:54uh very important area that is being
1:25:59talked about now how to execute machine
1:26:02learning deep learning algorithms on
1:26:05on these devices uh edge devices
1:26:08uh and and how that brings into a
1:26:12picture a lot of security issues um
1:26:16so uh given that it it definitely is
1:26:20an issue that iot devices are
1:26:23going to open up more
1:26:27security vulnerabilities but but the
1:26:31important aspect that needs to be kept
1:26:33in mind is
1:26:34that uh these are still distributed
1:26:37devices
1:26:38and although of course they are
1:26:40connected and if there are
1:26:42uh if there if some part of the
1:26:45uh network if some part of the system is
1:26:48connected to this
1:26:50iot device the attacker may be able to
1:26:53access some other devices in the network
1:26:56and then
1:26:57cause some issues but
1:27:00one thing that is being one thing that
1:27:03happens because of
1:27:05uh having these end nodes
1:27:08is that things are more distributed so
1:27:12it's more easier to handle the security
1:27:17issues that are there rather than
1:27:20things being centralized at one location
1:27:24and then and if some attack happens
1:27:28there then
1:27:29it becomes completely
1:27:32it creates issues in terms of
1:27:36addressing them the costs involved there
1:27:40and of course the
1:27:42time taken to handle that
1:27:45so so if if there are mechanisms in
1:27:49place
1:27:49in the iot devices
1:27:53such that or the network of iot devices
1:27:56such that
1:27:58things can be localized uh
1:28:01even even if some attack happens in some
1:28:04device uh in a in a particular location
1:28:08uh the uh the
1:28:11idea is that the
1:28:14propagation of that to a large part of
1:28:17the network
1:28:18uh can be limited yeah so uh
1:28:21of course this is a this is not a uh
1:28:25this is not something that uh
1:28:28has been done and dusted
1:28:31there are there there is still a lot of
1:28:33work that needs to happen in this
1:28:35scenario uh because as you mean as you
1:28:38rightly mentioned the
1:28:40iot devices have their constraints in
1:28:42terms of resources
1:28:44so we have to uh take those resources
1:28:47into
1:28:49those resource constraints into a
1:28:51picture while
1:28:52designing appropriate security
1:28:55mechanisms
1:28:56so there are some there are some
1:28:58lightweight
1:28:59security mechanisms that are coming into
1:29:02uh you know people are trying to develop
1:29:06even to you know run in
1:29:10things like uavs
1:29:13and because uavs also have
1:29:17issues in terms of resource constraints
1:29:20but but people are coming up with the
1:29:23lightweight mechanisms
1:29:25and that will try to
1:29:28limit the attacks
1:29:31that are possible and also i think there
1:29:34needs to be some
1:29:37concerted effort in terms of uh
1:29:40having mechanisms that will limit the uh
1:29:44propagation of the attacks towards more
1:29:46number of devices in the networks
1:29:48yeah but that's that's an interesting
1:29:50question and of course there is
1:29:52no uh no
1:29:55one answer to it i think there still is
1:29:57a lot of work
1:29:58that is that needs to be done in that
1:30:00area
1:30:01uh sure sir thank you uh just one more
1:30:04question can i ask
1:30:06sure yeah please yeah so uh it's also
1:30:10related so when we talk about
1:30:12cloud i personally as i understand
1:30:15whatever i've read
1:30:16is that whenever you you plan to use
1:30:20cloud computing you are
1:30:21actually trusting the service provider
1:30:25to a great exchange so when that trust
1:30:28factor comes
1:30:29in the question comes about privacy
1:30:33so uh if we are having a network let's
1:30:36say iot devices are connected and then
1:30:39uh there's an edge layer where gateways
1:30:41are there and then we are finally
1:30:44connected to the cloud somewhere so uh
1:30:47the question over here is that uh when
1:30:49we are blindly
1:30:50trusting the uh like i'm saying blindly
1:30:53trusting because we are giving a lot of
1:30:54access rights
1:30:56they can access our mac addresses they
1:30:58can identify our devices a lot of
1:31:00privacy is
1:31:01compromised when we trust that
1:31:04particular service provider
1:31:05so the question is that is there a
1:31:08mechanism
1:31:08in which the user or the person who is
1:31:12deploying the network the iot network
1:31:14has some sort of privacy control
1:31:17mechanisms available
1:31:18because if everything is accessible to
1:31:20the
1:31:21service provider you are highly
1:31:23vulnerable in terms of your privacy
1:31:25being compromised
1:31:27yes so of course privacy
1:31:32there are there are a lot of mechanisms
1:31:36that are
1:31:38exploited where uh
1:31:41we can try to have uh you know privacy
1:31:45is a lot of privacy information
1:31:47concealed
1:31:49um so there are various mechanisms
1:31:53in in the area of privacy that
1:31:56people have tried to or people are still
1:32:00working on
1:32:01it's it's still a very ongoing
1:32:04field of research so
1:32:07there's a lot of work in this area and
1:32:10privacy definitely
1:32:11in term for iot networks uh even even if
1:32:15i
1:32:15uh if if i may say there are middlewares
1:32:18uh that are deployed
1:32:21which have uh i mean base
1:32:24these are standard based for iot
1:32:27standards
1:32:29and they have
1:32:33some privacy control mechanisms that
1:32:36the middleware implementations need to
1:32:39have
1:32:40uh so that uh there are accesses
1:32:44given uh you know based on
1:32:48uh the uh based on whoever is
1:32:52trying to access and um and based on the
1:32:56permissions and
1:32:57uh if if uh
1:33:00if there is some uh data that needs to
1:33:03be concealed then
1:33:04then uh that that that is something that
1:33:07is
1:33:08uh taken care of so so yes there are uh
1:33:11there are some uh mechanisms that are
1:33:14happening both from the
1:33:16i think privacy algorithm point of view
1:33:19and also
1:33:19in terms of these middleware layers that
1:33:23are
1:33:25coming up in the iot standards
1:33:30domain

About this transcript

This page contains the full transcript of Cloud, edge and fog computing for IoT: Part I by NPTEL - Special Lecture Series, generated from the public captions YouTube serves with the video. The transcript has 11,499 words across 2,116 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.