YT YouTube2Text

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

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

by NPTEL - Special Lecture Series · 6,016 words · 1,124 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:03okay so
0:04um we the last
0:07session we left at this slide and we
0:09motivated uh
0:10that there are some issues in cloud
0:14computing
0:16that do not go well
0:20with the iot scenario
0:23and volume latency and bandwidth being
0:26one one of the main
0:28i mean few of those uh aspects that are
0:30really
0:32making it difficult with the increase in
0:34the number of
0:35connected devices uh so um
0:40what is the solution to it uh it is
0:43nothing but uh
0:45edge computing uh of course one one
0:48solution to it is
0:49uh that you keep increasing the uh
0:53data center size uh but as you saw in
0:56in one of the figures there
1:01for for the kind of uh for for the
1:04number of devices in several
1:09billions uh the the amount of data that
1:12it generates needs about
1:14uh several thousands of
1:17data sensors right so that is that is
1:19very costly in itself and uh
1:22uh therefore uh there needs to be a
1:25different way
1:26to solve this problem and uh
1:30one of the problem one of the solutions
1:32that came up
1:33uh is this fog
1:36or edge computing so we'll generally
1:40call it edge computing and uh we will
1:43see
1:44uh there there are various flavors of
1:47edge computing
1:48uh how how these different
1:52edge computing uh terminologies
1:55uh i mean the the different kinds of uh
1:58edge computing that we'll
2:03the classifications we'll see how they
2:04are different so
2:06edge computing is
2:10nothing but
2:11[Music]
2:13architecturally it is not
2:16a very complex thing because your
2:20what you are essentially trying to do is
2:23instead of
2:24having this centralized node or the
2:27centralized
2:28data center that you have uh
2:31instead of that you are trying to have
2:34some
2:35other set of nodes
2:39not just one but distributed
2:42in the space between the cloud and the
2:47end device and you have several such
2:51nodes and
2:53these nodes are going to be
2:56much more closer to the end device
2:59rather than how the current cloud
3:03is situated so it's a dist this is uh by
3:06definition this is a distributed
3:08uh information technology architecture
3:11uh where the client data that is
3:14the data that is generated by the end
3:18device
3:19is processed at the periphery of the
3:22network so
3:24you're not coming to the core of the
3:27network where the cloud exists but
3:28you're
3:30processing it at the periphery of the
3:31network uh and
3:34if possible very close to the source
3:37where it is originating
3:39yeah so um
3:42so one thing that definitely
3:46is problematic as we also discussed in
3:49the last session was um although you're
3:53bringing all that
3:54[Music]
3:55you're introducing new nodes um and
3:58which are closer to the
4:00uh originating uh end device uh data
4:04originating end device
4:06uh it these nodes are uh
4:09going to be less capable compared to the
4:14uh compared to the compute capacity
4:18and different other kinds of capacity
4:19that
4:21the cloud would have so
4:24so that is one problem that we have
4:28with these edge nodes but we have we
4:30have to see how we can
4:32very efficiently uh do things that we
4:35could do
4:36uh with the cloud uh while
4:40uh you know in remaining within the
4:42constraints of
4:44what the edge node provides um
4:47so the advantages of course
4:50by introducing the these new uh
4:54nodes uh between the a cloud and the
4:58end device is that
5:01all the data that the sensors collect
5:07they are very securely uh processed in
5:11real time right okay so um
5:15the the reason it's uh uh
5:18considered to be more secure is because
5:20your uh
5:21the the uh
5:25chance of the data that is getting uh
5:27generated
5:29uh being uh tampered with
5:32or being uh or the
5:36data being uh you know manipulated is
5:39very less
5:40because you're you're consuming it
5:43or you're working on it at a very closer
5:46location than the
5:48uh cloud so uh
5:52there is the possibility or the
5:54probability of that data getting
5:56uh getting uh
5:59you know uh into wrong hands is uh
6:03that much more lesser so that's that's
6:05one
6:06uh reason why it's
6:09also said that it is secure
6:14so uh other other important thing here
6:17is that it is done in real time because
6:19now
6:20you're not you don't have to
6:23go all the way back to the cloud
6:26do the processing and then come back
6:29so so that round trip time is
6:33really minimized
6:36to a large extent and therefore the real
6:40time applications are served better by
6:43this kind of new infrastructure that we
6:45have
6:47okay uh are there some questions
6:50no okay
6:56yeah so um so what is the
7:00uh what is the reason we had we find
7:03this need of edge computing
7:05uh the the problems that we saw in the
7:08uh
7:08cloud computing model right so uh
7:11one of the challenges was latency
7:15so edge computing addresses that problem
7:18as we just saw by decreasing the latency
7:21in
7:23transmission and
7:26[Music]
7:28i mean so so to say even for processing
7:32maybe it's not
7:33really uh decreasing the time for
7:36processing because
7:37cloud may have uh a lot more processing
7:41resources than
7:42what an edge node would have but
7:48in effect if you see the transmission
7:50time
7:51is quite large and that's why
7:54when you save on the transmission
7:56transmission time
7:58and with the with a minimal set of
8:01resources that you have
8:03on the edge it would still be better to
8:07compute the process the data
8:11on the edge uh but but there are there
8:14are some trade-offs there
8:15so uh that's something that we will have
8:18to
8:18uh you know consider while while
8:21uh deciding whether the data needs to be
8:23or whether the processing needs to be
8:26done on the edge or the clock so one
8:28important thing that
8:30needs to be remembered here is
8:33even though we are putting edge
8:37nodes or fog nodes let's say within
8:40uh in in between the space of uh
8:44the cloud and the end device uh we are
8:48not eliminating the clock
8:50yeah so cloud will still remain uh
8:53it is still there for uh
8:56uh for a purpose of uh you know having a
9:00uh bigger picture of the
9:05set of uh bigger picture of the
9:07environment and
9:10the the set of nodes that are there and
9:12the
9:14set of applications the models may be
9:17if you are running some algorithms that
9:20are using
9:22machine learning and all that so there
9:24are there are a lot of things that the
9:26cloud can still do
9:28um and therefore cloud does not go away
9:31the only thing is the burden of
9:33cloud doing everything is
9:36taken away and a lot of it is
9:40now pushed towards the edge of the
9:42network
9:44to the edge nodes so that's the
9:48main thing that you do and thereby
9:51latency is diminished a lot and
9:57the amount of data that is that needs to
10:01be transmitted
10:02is uh you know reduced
10:06by large orders of magnitude because
10:09uh currently it can also happen that you
10:14you can do some very
10:17intelligent but
10:19[Music]
10:20not so expensive processing on the
10:23edge device and
10:27come to a conclusion that not all of the
10:30data that is
10:32sensed by the various sensors
10:35or the data that is
10:38communicated by the end device
10:42needs to be sent to the cloud so there
10:45is an in
10:46initial filtering that is employed
10:50uh in the end nodes uh which make it
10:53uh uh you know very uh important
10:56because uh uh it initially all the
11:00raw data was sent to the cloud which now
11:03does not need to
11:04be sent uh because that there is this
11:07intermediate node which will
11:09take care of it okay so uh that is one
11:12uh
11:14advantage that also is brought about by
11:16the
11:17edge computing so of course
11:21the three issues that we saw in the
11:24cloud computing scenario was there was a
11:26large volume of data
11:29bandwidth was an issue and latency was
11:31an issue so you see that
11:32latency is solved the volume of data is
11:36handled
11:37by the intermediate node and
11:40there is only a little uh there is only
11:43a
11:44fraction of that data that is sent uh to
11:48the cloud
11:48uh thereby uh reducing the uh
11:52transmission of data also uh so that's
11:55so these are the advantages that are
11:57brought about by
11:58uh introduction of the uh edge nodes
12:01uh of course it means that you are
12:04now uh supposed to have this
12:08additional set of nodes
12:12and there needs to be
12:14[Music]
12:16mechanisms to manage these nodes
12:19but the the advantages
12:22outweigh the the
12:26challenges that it brings it brings with
12:29it
12:30and therefore although there are
12:33challenges
12:34those challenges are being worked upon
12:37by the
12:38uh research community and
12:41industry as well so uh so uh
12:44that's that's something uh very uh
12:47good about edge computing and uh also
12:50there are
12:51some open uh forums uh
12:54like open edge consortium uh which
12:57which are trying to uh
13:00come up with some um you know standards
13:03uh in how these edge nodes need to be
13:08uh so what what needs to be the
13:11characteristics of
13:12edge node deployments um
13:17what kind of services they need to
13:20provide
13:21and what kind of applications need to be
13:23considered
13:25so the use cases all of these things are
13:27worked upon by
13:28uh the edge open edge open fog
13:31consortium
13:33uh so uh that that's something that
13:36you can actually go and look at if you
13:39are interested in
13:40what the open for consortium is doing
13:42you can just go and
13:44look at it their website things are
13:48available lot of white papers and all
13:51are available
13:52uh so there are other consortiums also
13:54which work in this area
13:55uh uh we'll get into one of them uh
13:58briefly in one of the slides later
14:01so
14:05there are so edge computing is one very
14:09general term of this kind of a
14:14model where you are now distributing the
14:17computing
14:18uh across multiple nodes
14:21um and it's not centralized in the cloud
14:24uh but there are uh different
14:28ways of uh doing this and
14:31there are three models that are there
14:34that represent the edge computing uh
14:37paradigm
14:38uh one of them we is called fog
14:40computing
14:42uh then there is uh something called
14:44mobile edge computing
14:46and uh there is also a model
14:49that's called cloudlet um
14:53so all of these uh you know have uh
14:57they are relevant in certain
15:00scenarios uh but you can see that
15:04all of them try to limit the bandwidth
15:08limit the latency requirements
15:12and try to minimize the network
15:14congestion
15:19so if you look at it the characteristics
15:22of
15:25edge computing uh there are uh
15:28so so some of these aspects that we
15:30looked at that it tries to do better
15:33uh in comparison to cloud computing
15:35feature here
15:36uh so uh you can see that
15:39[Music]
15:40there is a low lower pressure on
15:42bandwidth
15:44uh so the bandwidth between the
15:47edge of the network and the
15:50core of the network where the cloud
15:53resides
15:54is is reduced dramatically by
15:58bringing in this these set of edge nodes
16:01or
16:02fog nodes and
16:06that is a very uh important aspect uh
16:09also because you don't have to now
16:13transmit that much data and that also
16:17reduces the cost of data transmission so
16:20it's not only the congestion aspect
16:24due to limited bandwidth resources that
16:28are available but also
16:29the cost associated with transmitting
16:32that this amount of data that's also
16:35reduced and and
16:39of course the latency is is a very
16:41important aspect because
16:44as i mentioned in the session in the
16:47morning lot of applications they require
16:50uh they they have stringent timing
16:53constraints and
16:55in order to
16:58satisfy the the real-time requirements
17:01that that's that's one of the issues
17:04that needs to be solved which
17:06was an issue with the cloud computing
17:08model
17:09um then also one thing that we discussed
17:13earlier
17:14was the data security part so
17:18how does edge computing
17:21how does it provide better security than
17:24cloud computing so this was something
17:26that we touched upon earlier that
17:30since things are now more distributed
17:33uh and and uh and and since these nodes
17:36are much more closer to the
17:38uh to the end devices
17:42um an attack uh
17:46the chances of an attack uh are
17:49minimized are lesser because you are not
17:53sending the data all the way to the
17:55cloud and
17:57there are uh the data recites in a more
18:01localized
18:02scenario thereby it is much more
18:06easier to employ some
18:10security mechanisms also the distributed
18:13nature gives it uh
18:16a little bit more uh resilience to
18:19security attacks because uh the
18:23the uh attack on one of the nodes may
18:26uh not result in propagation of the
18:31propagation effect to other nodes so
18:34uh the that way also it is uh
18:38there can be some things that can be
18:40avoided
18:41in the edge no edge computing scenario
18:45in comparison to the cloud computing
18:48model
18:49uh and finally uh there is increased
18:52availability
18:53uh because now you have
18:56the set of distributed nodes
19:01instead of a centralized
19:05compute node although you know
19:07centralized node
19:09is also a set of nodes but
19:12it's not centralized it is more
19:16distributed so even if a particular
19:20node in the h computing model goes down
19:24due to some reason uh it still
19:27you know can function because the there
19:30are other nodes which are
19:32i mean the the end device can still
19:34access some other
19:36edge node which is in the uh coverage
19:39area of the
19:40uh in the coverage region so so that's
19:43uh that's that's uh why it's
19:47there is some increased availability
19:49that is observed
19:50and all of these make edge computing a
19:54very
19:55uh a very good solution
19:58to solve all the issues that we had with
20:02cloud computing of course with uh
20:06i mean uh this this comes with it with
20:09some
20:10uh issues and we need to also handle
20:13those issues right
20:15okay so um the
20:20uh the uh will we look at uh for
20:23computing and what fog computing is but
20:26uh the uh for computing
20:29uh platform uh this is a template
20:33platform that fog computing nodes
20:36uh typically implement and
20:42there are a set of aspects that are
20:47required to be there in a for computing
20:50uh platform uh which kind of addresses
20:54a lot of
20:58you know requirements that people the
21:00end devices or end users may
21:03uh have okay so so there are some uh
21:06things that need to be uh available
21:09within the fog
21:10node uh to serve the requirements of the
21:13user
21:14and uh to towards that there you can see
21:17there are some platform services that
21:19are there
21:20uh within the uh fog computing node
21:24uh and these platform services
21:27uh let's uh go through them uh briefly
21:31one by one so
21:32of there is an authentication
21:34authorization
21:35uh service so uh it it is
21:38uh ensured that every access to this
21:43is access to the fog node is done only
21:46by uh authorized
21:49users so so that there is no
21:54malicious user present and trying to
21:56attack the node
21:58okay so that is uh there is a security
22:01mechanism
22:03that fog node needs to support uh there
22:06are different kinds of services that uh
22:09fog node may provide
22:13so typically every
22:16end device may have a different set of
22:19services that it
22:20wants for example it may
22:24for some end devices uh
22:27it may be to download um
22:30a set of um you know
22:35download the map of a particular region
22:37let's say
22:38uh you're moving with your mobile
22:41and you want to download the map of a
22:44particular region
22:46in that case the service that it
22:48requires is this
22:49downloading of the map so there may be
22:53different
22:53kinds of such services may be listing uh
22:56the
22:57uh listing the set of restaurants in a
23:01place
23:03so if if a person wants a
23:07person carrying a mobile device wants
23:09that
23:10kind of service uh it can uh
23:14you know request that service from a fog
23:17node
23:17and the fog node uh
23:20if if it has that uh service uh
23:24uh in in the uh in the platform
23:27then it provides the required
23:30information back to the
23:31uh mobile device so then when there are
23:33different kinds of services
23:36[Music]
23:37you need to manage these services
23:39properly
23:41and service management is one of the
23:45functionalities that is part of the
23:47platform services
23:50then there is a
23:51[Music]
23:53functionality called or a
23:56block which is called the cloud agent
23:59uh so typically as i already mentioned
24:03uh
24:04a fog node is not some is not a node
24:06which is
24:07uh uh sitting uh uh
24:10you know uh isolated from the rest of
24:13the
24:14infrastructure right so uh the cloud is
24:17as
24:17and and as i already mentioned cloud is
24:20still a part of the
24:21uh whole hierarchy uh so
24:25uh there is some communication that
24:27needs to happen with the cloud
24:30and uh this uh thing uh
24:33is initiated by the cloud agent so if
24:35anything
24:36uh if if there is any uh communication
24:40that needs to be done uh with the cloud
24:44that that happens uh that that
24:46particular
24:47communication is initiated by the cloud
24:49agent
24:51uh of then uh there are some location
24:54services
24:55so uh one of them as i mentioned is the
24:59is the part where let's say you're going
25:03to a place and you want to know
25:05uh what are the uh
25:07[Music]
25:09uh one one example i mentioned is what
25:11are the restaurants that are there in
25:12that place so that is a location based
25:14service
25:15uh there can be various such other
25:18things like what if you want to look at
25:20a particular kind of
25:21uh
25:23[Music]
25:25garment shop in a particular region so
25:27some some similar kind of
25:28uh location-based service uh if you
25:33uh want uh uh you know location-based
25:36services like this
25:37that that's also something that is part
25:40of the platform services
25:44one important thing that i would like to
25:48point out here is there is
25:50something called as the offloading
25:52management and this is a very crucial
25:55feature within uh crucial functionality
25:59that a frog node
26:01needs to have uh this is what
26:04act this this is what that decides
26:07um how the how a particular
26:12uh functionality or a service needs to
26:15be
26:16executed for an end device so for
26:18example
26:19let's say i uh want to execute
26:23a function
26:27let me just call it a function x i just
26:29want to execute a function
26:31x now
26:36there are uh there is a way where i
26:38could execute the entire function
26:40x on my end device itself uh
26:43or if the end device is not capable
26:45enough to do that then i could send that
26:47function to the fog node here
26:51yeah so the fog platform here uh could
26:54uh allocate resources so before uh i get
26:57into that you can see here that there
26:59are these
26:59different uh vms right the virtual
27:02machines
27:03so uh even in fog nodes
27:06you would have different vms
27:10and basically apps can
27:14be given resources
27:17on these vms and they can execute there
27:21now so if if there is a
27:24service that our function that needs to
27:27be executed
27:29on the fog node there will be some
27:31resources allocated here and they will
27:33execute it now it could so happen that
27:36the
27:38because of the resource constraints on
27:40the fog node
27:43it might not be possible to entirely
27:46execute them on the fog node so in that
27:49case
27:50uh there needs to be a way to
27:53offload a part of this uh functionality
27:57to the cloud uh and and
28:00you you could execute a part of it on
28:02the
28:03fog node but uh some some other part
28:07which
28:08uh even if so there could be several
28:10parts in a
28:12function where some of it may be
28:15critical to
28:16uh you know so that the data is
28:19available
28:20uh soon but some part of it may be for a
28:23longer term
28:24and therefore you don't need to get the
28:27results
28:27as soon as you want as soon as
28:30the first part requires so so in that
28:33case then
28:34uh there there needs to be some
28:36management done where
28:38some part of it is kept here and some
28:40part of it is sent to the cloud and that
28:42is what
28:42the offloading management does uh
28:45so this is something very crucial uh in
28:48a fog complete
28:50computing platform because uh you are
28:53not entirely uh discarding the cloud you
28:56are you're still using the cloud
28:58for some very
29:01special purposes so um
29:05and then and then you typically have a
29:08communication service where
29:11which which allows you to send data
29:14or communicate data with
29:17various kinds of devices it could be a
29:20sensor node
29:21or it could be some device which is
29:25which itself has some hardware and
29:28os and applications running on top of it
29:32like the laptop and
29:35the mobile device and so on so it could
29:39be
29:39communicating with various kinds of
29:41devices and this communication service
29:43is
29:45handling that and there are certain
29:48set of apis which are available
29:51which access the underlying layers
29:56so you some of the functionalities that
30:00are
30:01there in the underlying layer are things
30:04like load balancing
30:06virtual machine scheduling network
30:08management system monitor
30:10uh so what uh um so if if you want
30:13so there are multiple uh uh things
30:17multiple functionalities that are
30:18executing on the upper
30:20layers here if you want to
30:25balance the load
30:29across you know multiple set of nodes
30:32or vms that are there so there needs to
30:35be some
30:36load balancing algorithm that's there
30:39and that will
30:41take care that it's not the the
30:43functionalities are not
30:46you know mapped to a specific set of
30:49resources that are there
30:50so uh also there
30:54there are no various
30:57virtual machine scheduling algorithms
30:59that are
31:00available which can also be used based
31:04on what kind of
31:07output requirements you have
31:11of course the network needs to be there
31:14is some network management that needs to
31:16be done
31:17and your system needs to be monitored
31:20uh to see if uh
31:23if if the resources are
31:27being if there are resources that are
31:30not being utilized
31:32uh efficiently or uh are there some
31:35resources which are being utilized more
31:38than what uh it what it should
31:41so uh there are there are various kinds
31:44of things that
31:45[Music]
31:46functionalities that are built into this
31:48and
31:50that that is running quite
31:53you know very very close to the hardware
31:55layer above the virtualization layer
31:59so a fog computing platform uh typically
32:02looks like this and
32:04you have these different layers and
32:07within these layers you have these
32:08uh functionalities that
32:12for computing no doubt to uh provide
32:16um so apart from that
32:19uh i think uh this this
32:23picture is kind of self-explanatory so
32:26uh
32:28it's it it's just showing you the
32:29hierarchy where there are these
32:31end nodes uh available uh and they are
32:35communicating to the
32:36fog node and via the fog node uh to the
32:40cloud node uh based on whatever
32:43is required from the cloud yeah so uh it
32:46of course it could so also happen that
32:48uh the
32:49for the end nodes are directly
32:51communicating with the cloud node for
32:53some
32:54uh specific purposes um
32:57so that that communication is still
32:59there uh
33:00but some of the major uh computation
33:05uh functionalities are first
33:08handled by the fog node and then
33:13some remaining part may be sent to the
33:15cloud okay or or some
33:17uh data might be filtered at the fork
33:19node and then the remaining data may be
33:21sent to the cloud
33:22so one of these things would happen
33:27okay so uh having looked at that
33:30and before going into the different
33:32models of
33:33[Music]
33:35the for different models of edge
33:38computing
33:39let us look at an example to see
33:42uh the potential of edge computing
33:46and and what better
33:50example can you have then if you can
33:53provide
33:53some efficiency in the healthcare
33:56industry right
33:57this is something that is very crucial
34:01in in various ways and
34:05it has its age
34:09uh you know age of the population that
34:12is
34:13criteria that needs to be looked at uh
34:16the emergency
34:19with which you address the requirements
34:22of
34:23a patient that is also one of the
34:27issues that you need to very carefully
34:31address so so there are multiple uh
34:34aspects with the healthcare industry and
34:37and the applications associated with the
34:39healthcare industry and
34:41uh edge computing plays a part in that
34:44so
34:45you can see here there are three
34:47different scenarios that are
34:48shown um
34:52so that one of the scenarios that you
34:56could have
34:57so i i'll go to the details of these
35:00scenarios in the subsequent slides but
35:03to
35:03just show you there are uh multiple
35:08uh locations where you can have the
35:10edges
35:12and uh it plays its own part yeah so
35:16you could have a device that is
35:19sitting at that that you that a patient
35:21has at home
35:23which could provide remote monitoring
35:26and care
35:28and that device takes uh some crucial
35:31decisions or
35:32you know monitors uh uh
35:35crucial decisions may not be taken uh at
35:38the device itself
35:39uh but uh you know it may monitor data
35:42and
35:43provide data in time so that the
35:47healthcare expert can then come back
35:50with
35:52what needs to be done so that is one way
35:55of
35:56using the edge but still that edge
35:59device has to
36:00execute some functionality okay so that
36:02device sitting in your
36:04house then there are on premise
36:08edges which means edges edge nodes which
36:11are
36:12in the healthcare promises hospital
36:16promises
36:17so it would have so there are multiple
36:20[Music]
36:23multiple devices that you see in
36:25hospitals which
36:26are used for patient monitoring
36:30so these are the on premise edge devices
36:35and then there are devices that you
36:37could have
36:38in a connected ambulance so
36:42that's that's going into the realm of
36:46an edge device uh which is not static
36:50right so you are in this case your your
36:53connected ambulance is moving uh but
36:56still
36:57it is something very important we'll see
36:59why
37:00okay so uh there are different types of
37:04edge devices
37:06of course but uh it also uh is important
37:10uh
37:11where that edge device is so the
37:12location of the edge device is uh
37:14important uh and based on that uh
37:19it does uh it does uh some
37:22uh functionality that
37:26is suitable to be executed in such
37:29platforms
37:32and uh and of course i mean this this uh
37:37spectrum is just showing you uh the the
37:41horizontal uh access is just showing you
37:45uh where the execution is happening
37:48the workload is being processed so in
37:51this case it's the end device
37:53uh in this case it could be the
37:56enterprise solution that you have
38:00with or an access point is there where
38:03you could send the data to the access
38:06network and then to the cloud probably
38:11and and here you're basically having the
38:15mobile node so there is some
38:18transportation that's
38:19involved and the mobile node may be
38:22communicating with
38:23uh within with another
38:27fog node or it could be communicating
38:29with the
38:30cloud of that cloud where the
38:34cloud of the particular hospital that
38:39that then sends data to the uh health
38:42care experts
38:43okay so so there there are multiple
38:46points
38:47where these edge nodes can be
38:50deployed uh that's the uh main
38:54uh aspect or the main carry
38:57[Music]
38:58main take away aspect from this slide
39:01here
39:06okay so let's look at these things a
39:07little bit more closely
39:09um so as i mentioned the connected
39:12ambulance
39:13um so one of the things that definitely
39:17uh is required which
39:20kind of lacks in the current scenario is
39:23that
39:24your ambulance is not equipped with
39:28um with the capability of
39:32monitoring the patient and at the same
39:34time
39:35sending the crucial vital informations
39:38or vital statistics of the patient to
39:41the
39:42health care expert so
39:46so as until the ambulance
39:49is you know reaches the hospital
39:52premises
39:53the health care experts are not aware of
39:56what is
39:57the patient situation okay so
40:00so that is one aspect that
40:04the connected ambulance tries to do
40:07so it keeps you know it has
40:12all the monitoring equipment
40:15there are some diagnostics that are done
40:18and
40:20by the time the patient reaches the
40:22hospital those
40:25very vital data are already available
40:29with the
40:30health care experts and they can be
40:32ready with what needs to be done
40:34or they can do an initial uh
40:38analysis of what the situation is and
40:42be ready with what needs to be done by
40:44the time the patient arrives because
40:46there
40:47there is this time between taking the
40:50patient from the
40:51home to the hospital which is crucial
40:54time uh
40:55often and so there if things can be done
40:58better uh during that time that that
41:00always is
41:02going to improve the chances of
41:06saving the patient so that is one thing
41:10but one of the things that we already
41:14saw what edge computing provides is low
41:16latency right so
41:17here we we do want things to happen uh
41:20we do want things to be processed fast
41:23and sent
41:24right so low latency uh in this case we
41:28also want mobility support
41:29right so we the node is not static at
41:32one place
41:33it is moving around in the environment
41:35so
41:36when it is moving around in the
41:38environment whatever services it needs
41:40should still be available to it
41:44the movement should not be a factor that
41:48causes the availability of service not
41:52to
41:53uh happen yeah so so mobility support is
41:56one thing that
41:58is crucial here and
42:01uh the uh some some initial
42:04processing or some uh monitoring and
42:07processing that's done
42:09um should be another the
42:12[Music]
42:14the ambulance connected ambulance should
42:16be able uh
42:17to do that with sufficient uh
42:20resources may not be too much resources
42:22but some
42:24some amount of resources to do that
42:26initial
42:27uh processing okay and
42:30then you can send the data
42:34over the network and
42:37that can help the health experts to
42:41to make their diagnosis and be ready
42:44with the treatment as soon as the
42:45patient arrives
42:48so uh it could also be that
42:51you want to do some live streaming of
42:54processed patient data so that's that's
42:57something that
42:58also uh this could enable
43:01um and and as i said the analysis part
43:05uh is is made more
43:09effective by doing this uh
43:13and and you could you could have at the
43:15edge
43:16itself some real-time diagnosis
43:19and uh some some diagnostic tools can be
43:24used to uh understand the situation of
43:27the patient so
43:28so all these kinds of possibilities are
43:31there
43:32uh and this is there within the realm of
43:35a mobile node
43:36uh which is traveling in the environment
43:38so so that's that's a moving edge
43:41node uh so to say uh
43:45of course after the patient comes in the
43:47hospital
43:49the the patient is monitored
43:52continuously
43:54and the monitoring devices now are
43:59so some you can see that in some
44:03situations these devices are standalone
44:06they are not connected uh
44:08or uh or these are connected
44:11uh and they are the data that is the
44:14large amount of data that's
44:15collected um
44:18i mean is either processed uh and then
44:23sent to a third party cloud or or you
44:26send the unprocessed data entirely the
44:28raw data to the third party cloud yeah
44:31so this is for a long term uh purpose
44:34right your monitoring this data and
44:36you're keeping a record of this data so
44:38this is a long term uh
44:40reason why you want this kind of data
44:43but
44:44one of the concerns here is that you
44:47you don't want the
44:50data to be available with um
44:54with with parties who want to utilize
44:58such health related data
45:00so some some security issues are there
45:04and some privacy concerns are also there
45:06so that needs to be handled and if you
45:10uh the the best way of handling that is
45:13if you have a private cloud
45:14uh on the hospital premise
45:19so there could be an on premise
45:22edge in the hospital that
45:25locally maintains the data therefore you
45:28know providing for
45:30privacy and whatever compliance
45:33is necessary and
45:39the of course the other thing is that
45:41this kind of connected
45:43set up in an in hospital
45:47platform in in inside hospital will make
45:50it easier
45:51for the clinicians and their
45:55productivity also increases because now
45:58uh
45:59data is available
46:02for them to go through quickly right so
46:04that is
46:06that is definitely one case of edge
46:10devices processing and
46:14providing the required information and
46:17finally we have this other extreme where
46:21you are looking at a remote monitoring
46:23and care
46:24situation uh so you the elderly
46:27population
46:29um uh you know they they are or even
46:33uh uh some uh population which is uh
46:37uh who can have
46:40uh some uh you know emergency situations
46:44if they want to be monitored uh closely
46:48uh then then this kind of remote
46:51patient monitoring based on
46:54edge devices could be a very good uh
46:58solution because uh because then
47:01uh the the healthcare workers can be
47:05getting constant updates on what is the
47:07situation of the patient
47:09[Music]
47:11of course i mean this these are again
47:14there are again privacy issues if you
47:16run these
47:17applications in the cloud and therefore
47:20it makes sense to
47:21run them on edge devices so
47:25sensitive data is protected
47:28by doing this there is there could be
47:31some
47:31initial very minimal processing also to
47:35be done to generate
47:36some alarm signals uh based on the
47:40conditions of the patient uh and uh
47:43that could be uh low uh it it
47:46typically needs to be a low latency one
47:49if it's a really
47:50serious alarm condition
47:53and and there again the edge helps
47:57and the of course the amount of data
48:00that
48:00needs to be sent based on the
48:04kind of monitoring that is done uh could
48:07also strain
48:09uh and lead to a higher cost strain the
48:12bandwidth and lead to higher cost which
48:14can be
48:14avoided uh by uh using the edge device
48:18yeah so
48:19so you can see the uh different uh
48:23places and locations where edge device
48:27plays an important role and and at
48:30and at each place it's it's a
48:33it's a different uh
48:37role that it plays so so there is some
48:39initial care that it plays
48:41when it it gives when it's doing the
48:44remote
48:44monitoring then there is some
48:47intermediate
48:48care that it take it provides when it's
48:52in the
48:54ambulance situation and and then
48:57there is this final place
49:00inside the hospital where you are
49:02constantly monitored and
49:05taken care of right so that so there are
49:08there are these various locations where
49:09it
49:10comes to use uh so having looked at that
49:14application that kind of gives a
49:18a really good view into
49:21what fog computing is
49:25or what edge computing is but what the
49:27various
49:28models could look like so um
49:31for computing uh this this
49:34uh terminology uh a fog came into
49:38existence
49:39um you know
49:43it's it's it's kind of a metaphor
49:45because uh
49:46you see uh that uh the there is this
49:49cloud-like
49:50fog that is uh very close to the ground
49:54right so uh basically that's
49:58close to the uh ground is
50:01kind of a similarity that you have
50:05for node which is close to the end
50:08device
50:08or which is at the edge of the network
50:10so that's why this
50:12fog computing terminology came about

About this transcript

This page contains the full transcript of Cloud, edge and fog computing for IoT: Part II by NPTEL - Special Lecture Series, generated from the public captions YouTube serves with the video. The transcript has 6,016 words across 1,124 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.