Re: [T2TRG] Comments on draft-hong-iot-edge-computing-02

[Alessandro Bassi <[e-mail]>]

Hi Yong-Geun,

Thank you for your reply.

In general, I understand you are making the point for a distributed
architecture, where processing and storage take place at the edge,
rather than a centralised one where all data is sent to a "server" and
processed there. Any architecture has pros and cons, and, while I do
agree that an exponential growth of connected devices will show the
limitations of a centralised architecture, what I am interested is to
see where the "boundaries" are. I will try to explain with an example. 

Problem: I want to install a camera system with facial recognition that
sends an alert in case of a known criminal is recognised.

Solution 1: I install N dummy cameras, I connect all of them with a
central server. The data stream is analised in the "central" location or
somewhere on the on the cloud where my services run.

Solution 2: I install N smart cameras, each of them with the
functionality of recognising a face in a set, and the camera will send
an "alert" signal if a face is found.

Both solutions work - on paper at least.

If I want to be very basic, I can see the two costs as follows:

Cost of Solution 1 is: (dummy camera cost) x N + connectivity to the
cloud + running the services on the cloud.  

Cost of Solution 2 is: (smart cameras cost) x N. + running some service
(mainly update pictures) on the cloud. 

It is clear that if the connectivity becomes a problem, it will become
very expensive (up to infinite if my requirements cannot be met) and
solution 1 is not feasible.  

Now, in real life the above equation cannot be just taken as-is: there's
the need to factor in the easiness of software upgrade in Solution 1
rather than 2, the fact that images need to be upgraded, which means
sent to cameras from time to time, that in the solution 2 there's the
need to send a signal if a face is recognised - which means, the cameras
must be connected at all times and there is therefore a connectivity
cost, albeit small, and so on ... but the idea is that if the
connectivity cost is bigger than the delta between a smart and a dummy
camera, then the distributed solution is better.

What it may be interesting is to study the characteristics of the two
architectures (or more, if you consider for instance computation at the
real edge - on the device - or in the local intranet) and look all the
different parameters that have an influence, to highlight which paradigm
should be chosen given a specific set of requirements. This for me would
fulfill "

 
"[...]the benefits and challenges of Edge computing mainly focused on IoT data.  The purpose of this document is to bring up the issues of Edge computing for IoT services in IETF/IRTF."

as you wrote. 

As well, please find some comments inline.

On 03. 04. 19 16:00, 홍용근 wrote:
>
> Hi. Alex.
>
>  
>
> Thanks a lot for your comments.
>
> With respect to all your comments on the lack of the evidence data or
> references, we will clarify them in the next revision.
>
> Other than that, please find the answers in-line.
>
>  
>
> Best regards.
>
>  
>
> Yong-Geun.
>
>  
>
> *From:*Alessandro Bassi [mailto:[e-mail]]
> *Sent:* Tuesday, April 2, 2019 7:44 PM
> *To:* 홍용근; [e-mail]
> *Subject:* Comments on draft-hong-iot-edge-computing-02
>
>  
>
> Dear authors
>
> I've read this draft and I also have a few comments:
>
> 1. Introduction
>
> "Nowadays, most IoT services are based on Cloud computing since it can
> provide virtually unlimited storage and processing power."
>
> Do you have any statistics to show that really most of IoT services
> are on the cloud?
>
> "about a half of data is stored, processed, analyzed and acted upon
> close to the data producer"
>
> Again, can you quote a source for this claim?
>
> [Hong] We think this is a common sense and it is easy to predict.
>
>  
>
Well, I'm sorry to insist, but if you say "around 50% of data is stored
close to the data producer", I would expect a scientific proof of this.
I'm not questioning the value - it could be perfectly valid. Indeed, it
can be common sense. But "common sense" made the people think for
centuries that the earth was flat, if you know what I mean ... so, in
general, I would refrain in indicating values if not supported by a
scientific study.  

>  
>
> 3.1
>
> In your definition, IoT does not have any actuation - do you really
> mean it? As there are dozens of definitions for IoT, I would recommend
> maybe to pick one of the existing ones.
>
> [Hong] IoT could include sensors, devices, and actuators.  We tried to
> provide very general concept of IoT since there are dozens of
> definitions as you said.
>
In my view, to be very generic, IoT MUST include a physical entity, a
digital entity (representation of the physical one, we can call it a
"digital twin"), a way to link the physical and digital world (through
devices, which can be sensors or actuators), and the possibility of
exchanging information. One rather poetic and extremely generic
definition I've heard for IoT is "where atoms and bits meet". The
definition you provide is

"the concept of IoT has been that things connected to the Internet can send and receive information collected by sensors without human intervention, where things are various embedded systems such as home appliances, mobile equipment, wearable
   devices"

I hope we can agree that this is not very generic ...

>
> 3.3
>
> "Now with IoT, we will reach the era of post-Clouds where
> unprecedented volume and variety of data will be generated by things
> at edge networks and many applications will be deployed on the edge
> netwoks to consume these IoT data."
>
> Is this your opinion?
>
> [Hong] Don’t you agree with it? Or are you asking the source as well?
> We will clarify this as well in the next revision.
>
It does not matter if I agree or not. "we will reach the post-Cloud era"
is a statement. If it's your statement, just say it.

>  
>
> "Some of the applications may have very short response times, some may
> contain personal data, and others may generate vast amounts of data.
> Today's Cloud based service models are not suitable for these
> applications."
>
> Why? 
>
> [Hong] Do you think that today's Cloud based service can satisfy any
> IoT applications which even requires very short response times and
> privacy? Or asking a source again? We don’t catch what you are asking.
>
>  
>
You cite three cases.

- very short response time: with the advent of 5G technologies that
_supposedly_ have a very short latency time, the response time should
not be an issue (or at least this is what the 5G narrative says). If you
disagree, please say why.

- personal data: it may be easier for an attacker to hack a device in
the wild rather than a well-protected computing center that is hosting a
cloud service. If a device is not able to encrypt messages in transit,
it will likely not be able to sustain an attack from a hacker.

- vast amount of data: if a device generates a large amount of data,
storing on-device will require that the device has a large amount of
storage, which may be expensive.


thanks! best,

--alex