Improving network energy efficiency with AI

0:03 [Music]

0:10 [Music]

0:25 hello you're watching The Green Network Summit part of our year round DSP leaders coverage I'm raila Matra

0:33 editorial director at Telecom TV uh today's discussion looks at improving

0:39 Network Energy Efficiency with AI it's one of the hottest topics in the Telecom

0:45 sector right now and there are lots of different views on how and when this might be achieved recent studies suggest

0:52 that energy costs account for on average about 23% of a telco's network operating

0:58 expenditure but can Telos use Ai and analytics to reduce costs in an

1:04 efficient way without significant investments in additional Network Technology and applications and without

1:10 impacting their services well during today's discussion we're aiming to identify some of the key approaches to

1:16 the use of AI Tools in improving Network Energy Efficiency and get a sense for

1:22 what strategies are worth pursuing and I'm delighted to say that joining me on the program today are Diego Lopez senior

1:30 technology expert at telica and an Etsy fellow merco volini VP of innovation at

1:38 cult technology Neil McCrae Chief Network strategist at Juniper Networks

1:44 and Beth Cohen Telco industry Analyst at Luth computer Specialists hello

1:50 everybody good to see you all there's a lot of ground to cover and I'm certain that our audience will send in

1:56 additional questions for our live show so let let's get started uh looking at

2:02 the planning and build stage of a network how can the use of AI specifically to improve Energy

2:09 Efficiency be incorporated into this particular process either for new sites

2:15 or upgrading older ones Neil let's come to you first yeah hi Ray so I mean look

2:21 this is a um kind of a bigger challenge you start starting with a framework

2:26 what's your framework to to look at plan and build and then it's it's it's it's largely the simplest of things which is

2:34 how do you design your core Network sites how do you design your tower sites

2:39 how are you planning the infrastructure throughout um your whole network what

2:46 advances in technology are you using in terms of cooling um how are how are you

2:51 using AI to actually help you plan the site so there's there's very many more

2:56 um AI based planning tools that will help with with airflow it'll help with uh designing the

3:03 power which which is all you know people when they think about Telco equipment they think you know routers and switches

3:09 and things like that but actually it's all the Power Equipment it's all the the air conditioning and there's a lot of AI

3:15 tools out there actually a lot of them come from the data center world where clearly they've got a probably a big

3:21 slightly bigger challenge than than we have in the Telco world but those tools are be imported into telecommunication

3:28 companies because actually some of the challenges are similar in terms of network nodes and network sites and then

3:35 I think the other thing which which we probably come on to later on which is ensuring that you've got good visibility

3:41 of of the infrastructure that you've got one of the biggest challenges I hear

3:46 from Network operators especially in the you know how do we optimize the legacy

3:51 is how do I know that I'm counting everything how do I know that I see everything um and that remains quite a

3:58 big challenge not just in this space but in in many other spaces but really engaging the you know pulling the data

4:05 from the network pulling the data from cooling pulling the data from energy usage trying to understand where you've got gaps um and then and then working

4:12 with these tools to to efficiently plan things and then also I mean we we even see some operators thinking about where

4:19 they locate their their sites um in terms of uh location of the of even the

4:25 buildings so that they they maximize on things like solar power or they maximize on on other environmental things that

4:32 help with running the the site more efficiently okay thanks Neil great point

4:38 there about uh inventory and these AI tools coming from the the data center

4:43 sector uh merco will come to you next and then Beth so uh uh merco how do you see the role of AI tools uh uh playing

4:51 in the in this Planning and Building process I think if you if you look at the level of complexity we have uh in in

4:58 networks and thinking about our Network we we operate a Global Network with hundreds of sites and and thousands of

5:06 links several hundreds of of sites it has become impossible to to do optimal

5:13 planning um and I think the the simpl use case the one we are adopting today is H for for AI is demand for casting

5:20 where you can actually use AI to predict where traffic is going to go and with that then you can overlay information

5:27 regarding where uh energy is more efficient is also Greener and combine

5:33 this information to ensure optimal deployment of infrastructure so I think that's applicable to both existing sites

5:40 as well as new sites but you can actually do this in a in um in um development uh forecasting type of

5:48 approach I think what what we what we see this evolving into tomorrow is more of like a digital twin type of model

5:54 where you can you can create a a duplicate of your network and uh using

6:01 these techniques as well as AI you can apply these forecasting techniques to war scenarios and with that then you can

6:08 simulate what would happen if you apply certain scenario what kind of energy

6:13 consumption would you have what kind of a level of uh Energy Efficiency you you're going to have again with multiple

6:20 scenarios so it's basically using AI as a predictive tool to provide the

6:25 insights into potential inefficiencies and optimize the the development yeah absolutely we're hearing more and more

6:32 about digital twins these days uh and Beth let's come to you um uh what role

6:38 are you seeing for AI in this network planning and build stage well i' I'd

6:43 like to pick up on some of the uh on something that Neil said about the the

6:48 um you know using um the cloud vendors um as kind of models um I think the

6:56 cloud vendors you know have it simpler actually because they tend to be concentrated in only a few sites they

7:03 can in fact pick and choose to be um you know next to a a power generating dam

7:10 for example uh so they can take advantage of um lowcost Power and um be

7:16 more energy efficient in that way because uh one thing is Energy

7:21 Efficiency equals cost savings so I think we you know that's that's part of the reason that Telos go after it and

7:28 and of course any industry um but we have it um you know the TCO industry

7:34 actually has it a little more difficult because uh first of all many of the sites have been around for decades and

7:41 and they've been optimized for other reasons um we also are running a a a

7:47 wide area network so sometimes the locations are set by things outside of

7:53 energy conf considerations um so that makes it just that much more complex um to to drive

8:01 that Energy Efficiency and I think that's particularly important at the edge where um where we have very you

8:08 know so then we at the edge we need to think about using um more efficient um

8:16 Hardware more efficient uh operations and and that's where we can

8:21 use AI to optimize at the edge um and you know obviously you can use it at the

8:27 core as well but I think it's going to provide more dividends at the edge okay

8:33 thanks Beth and Diego let's come to you next um what about this early stage in

8:39 the in the network rollout uh and even upgrader sites to to to bring down this

8:45 uh Energy Efficiency um overhead oh I I think that this is something that uh we

8:51 have to take into account as well is related with the fact that you have to

8:56 make room for the AI itself in the new deployment on when adapting existing ones because you you had to run and it's

9:04 you had to run if not necessarily the AI you had to run the data collection and

9:09 and some kind of data preprocessing to make it useful for the AI and in many cases the AI the AI is in themselves

9:15 because the the degree of automation that you can achieve is very is when it comes to energy

9:23 consumption is something that is should be quite local in the calculations and

9:28 and and the and with a with a short Loop for decision so so I believe that this

9:34 is this is an aspect that we have to take into account and rely on the fact that we we have the um the advantage

9:41 that with the evolution to of the network um infrastructure to more and more Cloud native virtualized Etc there

9:49 is the possibility of having a balance of where you run the functions and where you run the AI and you can plan for this

9:56 to be um to be uh well balanced and well structur without making a significant or

10:04 dedicated investment on on dedicated hardw work for for simple models that

10:10 would help you to to save energy at at a edge or a or a or a local point of

10:17 process probably you you don't need big uh big gpus or whatever but simply a

10:22 well-trained model and and and uh and and trustworthy data yeah no absolutely

10:28 you got to consider consider all the angles here in in whether you're actually making any gains by using AI in

10:35 this process uh and Neil did you want to come back in here with an additional Point yeah I mean and you know I agree

10:42 with um everything said I think the point I also wanted to think about was is kind of layers in the network we as

10:48 telecommunication operators have always add had these layers and in modern

10:53 networks with you know we've got 800 gig coherent Optics um that can enable you

10:59 take layers out and when you take a layer out that has a massive impact on your energy

11:04 consumption um so instead of having you know maybe three or four times the number of Optics you can cut that down

11:11 by maybe 50% maybe even more and we have customers at Juniper that are doing that

11:16 with our new PTX platform and and taking those layers out it doesn't also it

11:22 saves you not just in energy but it allows you to get upgrades done quicker faster and bring new Fe features to the

11:29 network so we see many Network operators um and we kind this term from one of our customers in the Middle East

11:35 doing this network modernization that's really using the best of breed capabilities that you can get today and

11:43 rolling them out in a way that makes a um a substantial difference in the whole

11:49 energy usage of the network and then the second point is is we talk about data

11:54 actually in this scenario most of the data is in our Engineers heads that are on these sites

12:00 and I really encourage everyone to go and you know as you're as you're building that data set for your AI

12:06 ensure that you're in you're interviewing and including the people who work in these sites day by day

12:11 because they've probably got the most juiciest knowledge about the site what

12:16 uses power what are the challenges that they face and that data is often overlooked because we're trying to pull

12:22 it from systems I really encourage people to go out and talk to the the on the ground experts in each of these

12:28 locations because they've got gold in their heads that if you extract it you can make a you can make a massive

12:33 difference in the modeling that you do okay yeah fantastic Point mind the M the

12:38 gold in in those heads uh that's that's what you want to do uh absolutely and and Beth will come back to you for the

12:44 final point on this question yeah I just wanted to pick up on what what Neil said about the the gold in in the engineers

12:51 heads um because it gets back to observability um which is if you're not

12:58 looking for that in the systems you're not going to find it and you're not going to get that information and that

13:05 information is critical and of course AI doesn't work if it doesn't have the information so yeah the engineers really

13:12 do know what's going on and uh yeah you definitely need to pay attention to what they what they know okay thanks Beth um

13:21 so we've just looked there at the build and planning stage but in terms of

13:26 network operations how is AI being used today by Telos to help improve Energy

13:33 Efficiency and to reduce Network power usage uh dieo let's start with you we we

13:39 we I think that uh well at least uh the uh let's say TI one tier one operators

13:47 are are applying it not massively and probably not in a substituting human the

13:54 human Loop the the AI is being used for sure for planning m was mentioning this

14:00 idea of demand forecasting we are using this uh currently and uh it's something

14:05 that is being used and in some cases we are achieving some time ago I had the opportunity to work with to work with

14:12 some colleagues that were making a um um work on AI to optimize not to optimize

14:19 the use of the network but to optimize the amount of energy that we were spending with the with the track roles

14:26 when attending um uh dealing with incidents at customers premises and it

14:34 was a it was an amazing amount of uh of uh of energy that we were saving invols

14:40 probably of fuel for the for the trucks and the uh and the Vans but it was we were saving energy as well so the idea

14:47 is that uh I think that right now we have we or we we have ai systems that

14:53 are being applied here and there for different uh for different activ

14:59 activities that are mostly uh focused on on assessing uh decisions that are taken

15:06 by by engineers and planners still we we still like the the uh point in which we we take the

15:13 follower the following step of making making AI more part of the U of the control at at the top of the of certain

15:20 control loops I'm trying to understand this and something that we have studed as well and and I believe is quite

15:27 interesting is trying to mix several different goals in the in

15:34 the AI or using the coordination of different AI modules to consider for

15:40 example that well you can you can achieve a particular uh very high energy

15:45 saving with a certain decisions on on routing or on on the uh on the devices

15:52 of of the configuration of the devices are active but at the same time you have to evaluate which is the the uh

15:59 impact that has not only uh in particular on the network stability on

16:05 the network characteristics as as a whole but on the U on on the on the user

16:10 experience in which AI is is again an an un valuable uh source of of predictions

16:17 and there are some colleagues of mine that are working on this and they are achieving a quite interesting results on

16:23 the balance between savings and the and keep while keeping a a good uh user

16:30 experience yeah absolutely a vital consideration there uh merco we'll come

16:36 to you next about the the current use of AI in energy optimization and then we'll

16:42 come to to Beth so uh Mera let's hear from you yes so I like to mention a

16:47 specific use case which is the the one of optimized routing um I mentioned ear

16:53 we have a very complex uh Network and actually it spans multiple countries and you I'm sure everybody is aware of the

17:01 price of energy is being heavily dependent on the top of uh power plants

17:07 you have and it can vary quite quite a lot country by country there actually some very large differences sometimes is

17:14 twice as much expensive kilowatt per hour in a certain country compared to another one so we have we have and again

17:21 with our complexity we have multiple options from go going from A to B for setting up netor routes for ourselves

17:30 for internal use I mean operations and also exposing that to customers so we have the ability to create a green paths

17:37 that doesn't necessarily mean the shortest possible path but but is

17:42 actually the the most energy efficient path it may not be again necessarily the

17:48 lowest possible consumption of energy but more the actual again lowest possible spend based on on on the path

17:55 that you choose so that's something we can um uh used for our own internal

18:00 optimization and also exposed to customers we actually built into our Nas

18:05 platform the ability for customers to see and choose which path they want to to to go through uh based on the metrics

18:13 they see about energy consumption okay great thanks merco

18:18 great to hear real use cases coming through now uh and Beth uh you've been

18:24 uh looking a lot into um the use of uh AI tools to optimize Energy Efficiency

18:31 at the edge of the network I understand yes um so I'm currently working with a a

18:37 group with the open infra uh Foundation we're uh The Edge working group we're working on a white paper on this topic

18:44 uh should be out in a couple months um but uh it's uh so what what is

18:51 immediately apparent um with uh you know with our initial conversations is just

18:56 how complex the issue is and um I think it's a good use for AI um because

19:04 there's multiple variations and variables uh and factors you know as as

19:09 merco and Diego both mentioned uh that go into uh optimizing the network for

19:16 Energy Efficiency but also balancing that with uh you know making sure the

19:22 network is available um at the same time I mean uh a lot of um data centers um

19:28 you know uh Cloud applications optimize their Energy Efficiency by um tracking

19:33 the usage and then just turning turning machines off for and then spinning them up uh when they're needed but uh Telco

19:41 usage um doesn't really follow that same very predictable path uh so it it

19:48 requires a lot more um understanding of the variables and uh that's where you

19:54 know AI can help um with the decision-making process and of course

20:00 you know as as uh uh Marco mentioned you know price of energy change is quite

20:07 variable across different countries you know political factors come into to play

20:12 as well uh where one country may be um pushing one type of energy over another

20:18 so you know the complexities are just just multiply um and uh you know that's

20:25 where I think AI can really help okay yeah absolutely um nothing is ever easy

20:31 it seems uh in this industry at all uh and Neil let's come to you uh finally on

20:37 this uh about the role of AI in helping with any Energy Efficiency in current network operations yeah so look you know

20:45 operations team their goal is to minimize change because change is typically what causes outages so I see

20:52 lots of our customers using AI on things like configuration management so we see

20:58 many times you know customers saying hey you know our power utilization is high

21:04 and we we we go and look and and do some audit work and we find that often

21:09 customers haven't turned those power saving features on and that it that that

21:16 in itself is something very simple to to fix very simple to do and we're encouraging our customers to to use AI

21:24 as a way of managing configurations and changes and but also looking at how we

21:29 optimize those configurations to get the most out of the network um and then kind

21:34 of second and and related to that is how do you how are you

21:40 ensuring that the the you know the data is in the AI model and and how are you

21:46 ensuring that your inventory is 100% up to dat I mentioned this earlier but increasingly we find you know the AI is

21:55 comparing the bill and the usage compared to what it knows about the site and it will flag hey there's something

22:02 missing here because in theory this is what the bill should be but the the the bill from the power company is this and

22:09 or the usage is this um and in ensuring that you that you kind of go back and

22:14 pick up on that because that in itself usually points to something that might be Legacy that actually you don't even

22:20 need and you can turn it off and the simplest way of saving money in many cases is what what are the things that

22:27 aren't that are there that we just aren't using anymore but we just haven't gone around and turned off and again an

22:33 operator here in Europe had a a kind of a hackathon that was discover things on

22:38 the network that weren't doing anything or weren't weren't generating any value and then just go turn them off um now

22:45 there's some risk to that so you have to you know follow process Etc but you know that can you know one round of that can

22:51 save you more than than any other program you've got because it takes time to deploy these things so really um

22:56 encouraging that kind of AI based configuration management um to ensure

23:02 that what you've got out there is what you should have there and anything that is that isn't there or anything that you

23:07 don't know about that you're flagging that and you've got someone looking into it okay yeah great points Neil I mean

23:14 this accuracy of data as a starting point is absolutely critical and as you

23:19 point out these tools can sometimes highlight where there is a lack of that

23:25 accuracy um well let's move on now because uh another thing that um some

23:30 industry organizations and uh operators and vendors have been looking at for a while is how to measure Energy

23:38 Efficiency um so is there a role for AI in helping with metrics and measurements

23:44 of energy usage to help Telos better understand the data uh Neil let's start

23:50 with you um I think it's a tough question right I don't I don't see anything coming out of the you know the

23:57 world that's saying you know this measure um is a sign of

24:03 you're doing things right now obviously we have all the the measures that are being driven by government in terms of

24:09 type one type two Etc and they're they're they're helpful in terms of thinking about the overall program that

24:15 you've got but I think I think we're still at the early days particularly when it comes to AI um you know in terms

24:22 of what those what those standards are one of the things that that we're working on at junipur is actually in the

24:29 in the routing standards so you know today's routing is very much focused on

24:35 shortest path um and you kind of building on what Merkel said is sometimes shortest path isn't the best

24:41 path from our energy utilization or even cost perspective usually it is with cost

24:47 but but but sometimes it isn't and what we've been working on is is how do we expand those standards so that we're

24:53 measuring we're taking uh you know a new parameter which is the the energy

24:58 utilization or the the green factor of any network Choice and then using AI to

25:05 kind of really go out and measure that say actually is this is this the best path for the network and one of our

25:11 platforms Paragon enables you to do that kind of visualization um and as we as we Define

25:16 and work with industry on those standards I think we'll see many more measures that that we'll that we'll

25:23 start to we'll start to understand and align on but I think right now it's we're probably at the the start of that

25:30 Journey um but lots of people across industry including ourselves are working on that okay interesting uh Beth are you

25:39 seeing uh any particular developments here using AI to measure energy usage uh

25:45 in networks um I'd say it's uh at the early

25:50 beginnings of it and and I think partially it has to do with that um the

25:55 biggest Energy savings is in the hardware um and that has not been a

26:02 priority of the hardware manufacturers and obviously Neil uh Juniper is clearly

26:07 getting the message um but o you know that uh traditionally it has not been a

26:14 high priority um for the vendors and you know I think now that there is more of a

26:20 priority um I think we'll start seeing better tools that will you know again

26:26 gets back to observability if you're not looking for it you won't see it um and

26:33 and so I think that there's lots of opportunities for AI to be um to be

26:39 applied uh but we need to to pull the data to get the the data um you know AI

26:45 relies on huge amounts of data and if we don't have the huge amounts of data yet which I don't think we do um you know I

26:53 think AI has limited um use um but that I think that's going to change in the

26:59 next 12 to 18 months as as we put uh higher priority on improving the

27:04 efficiency of the hardware itself improving the software that runs on it and in and adding observability you know

27:13 adding ways of observing uh what's actually going on in the

27:19 network okay great thanks Beth uh and Diego let's come to you I believe that

27:24 using AI some of one of the real challenges is here is not only about running the network but is about being

27:32 able to evaluate which are the pr price that you're paying in terms of energy

27:37 consumption when when providing a service uh where and and in that case

27:42 what you have is a big huge problem of data aggregation and of of identify or

27:49 separating data that is in many cases provided as a as a whole if you have a

27:55 big a big optical switch uh that is uh well working on 100 gigs or several 100

28:02 gigs Etc you have a lot of services that are going through and and you would like to

28:11 understand how which is the the composition of that energy consumption

28:16 how can certain certain actions may impact one service or the other on the

28:21 one hand on the other hand is to make your customers aware because this is an information that they are they keep Bing

28:28 and as the concern about the energy prices on the one side and all the concerns about the excess of energy

28:35 consumption of becoming greener in general as a society grows uh the uh the

28:42 idea of providing accurate um information so so people know what they

28:47 are what what they are consuming how they are consuming it is something that is important in this identification of

28:53 the data sources in this separation of data agregates Andre agregation in in

28:59 other in other U uh in another packaging that uh that

29:05 would be useful for for users to make informed decisions I believe there is a a great um field for for the application

29:13 of AI and this is something that well Neil mentions stand mention standards

29:20 before uh for example we have recently started an activity in the ATF that is

29:26 has a very original name it's called green working group uh in which that we are trying to identify which are the

29:33 data how how a data can be aggregated how can be assigned in in in temporal

29:39 units Etc and this is something that uh we are um we we are dealing with to

29:46 better understanding the uh what we need and where we can apply AI okay great thanks Diego the green

29:54 working group you know it does what it says on the tin so that's as good a name as as any I think right there okay uh

30:01 thanks everybody for that um now let's have a a broader look at the ecosystem

30:07 here because um you know there's a lot of uh expertise uh out there beyond the

30:12 Realms of the Telecom sector so uh how can Telos look beyond their usual

30:18 ecosystems to create new Partnerships that use expertise from Academia data

30:25 Sciences energy companies AI specialist and so on is there a new AI based

30:31 ecosystem emerging that will help with the green network uh merco let's come to

30:36 you first on this I think has been said already but one of the biggest challenges we have here is data

30:43 collection getting good quality data so I think the area where we can leverage

30:49 expertise that is typically not something that TCO sector has been leveraging a lot in the past is is this

30:56 area if you if you look at how we been operating networks typically data has

31:02 been collected and then thrown away so we're not we haven't kept uh good

31:07 historical data across the board and I think there is a a treasure of data that we can collect so

31:14 leveraging tools and systems and expertise in this space I think is the biggest opportunity for us okay thanks

31:21 MCO and uh Beth how do you see the ecosystem widening if at all oh I I

31:29 think it has to widen because uh you know Telos are really good at running networks um but Telos traditionally have

31:37 have mostly been consumers of energy um and yes there's been some energy you

31:42 know Energy Efficiency initiatives um but a lot of a lot of the

31:49 you know our relationships with the energy companies have been for the most

31:54 part hey how much are we going to pay for electricity um and I think that we need to take a a

32:01 l broader view um of because the energy companies know how to produce energy

32:07 they're looking for new ways to produce energy um and be more efficient about it

32:13 and lower costs um and and you know we we need to be there with them and and be

32:19 in partnership with them um because we are obviously one of the major consumers of of their energy that they generate

32:28 um I think another area that we should be looking at um you know standards bodies obviously is important um and

32:36 standards bodies not just the Telco traditional standards bodies that we that we normally work with um but you

32:43 know the the standards bodies that that are part of the data science standards bodies the energy um utilities standards

32:52 bodies as well as Academia I think there's a lot of opportunity um that we

32:58 miss out on Academia uh I don't think the Telco industry traditionally does

33:03 much with Academia um but you know as part of this this white paper I'm working on I discovered there's a whole

33:10 lot of people working on um uh Energy Efficiency in in Academia um you know

33:19 writing um you know their thesis and doctoral dissertations on this topic and doing you know great in great um work in

33:26 the labs to to really optimize um Energy Efficiency and and

33:32 let that drive um the decision-making process um and uh which is is again not

33:39 traditionally how Telos approach things um so I think that um we should really

33:44 take advantage of that broader ecosystem and and you know so that we can optimize

33:51 the delivery of our services um both cost effectively um energy efficient L

33:58 as well as optimally Optimal Performance okay thanks Beth uh and Neil

34:05 how do you see the ecosystem um spreading in this regard um I don't see it spreading hugely r i mean I think you

34:13 know everyone in in the kind of digital world is bringing on AI experts and I

34:19 think it's about you know pairing your AI experts with your network engineers

34:24 and doing that kind of self-learning shared learning process so that the tools and the systems that you're

34:30 building are actually driving the right value one thing I would say about the

34:35 ecosystem that's there today but probably doesn't have the right um priority in terms of management Focus

34:42 time which is every Telco spends Millions probably billions across the

34:48 whole industry on Power and cooling infrastructure um how many Telos look at

34:53 that as a Strategic investment and I can tell you the answer not many but actually the way that you are managing

35:01 that power build the way that you are managing that cooling build can give you a massive uplift in Energy savings so

35:09 rectification going from kind of Legacy rectification to sinewave based rectification that'll knock maybe four

35:16 or five percentage points in efficiency but you've got thousands of those devices in the network and if you're

35:22 thinking about that spend strategically and working with the right Partners probably partners that you've already

35:27 and you've probably got people buying on the dra trying to sell you the next thing in this space but it's kind of

35:33 lost in in the in the kind of dungeons of the company because it's it's not the

35:38 the the most coolest stuff like AI that you're working on but let me tell you those areas um and those companies are

35:45 leveraging AI today as well they can have a massive benefit and then how those organizations work with your other

35:51 vendors so that if we're bringing in you know a a new MX plat platform or we're

35:58 bringing in a a a server rack how are you working with all of that ecosystem

36:04 to ensure that you're doing that in the most efficient way what I see is quite a lot of Silo thinking in Telos whereas in

36:11 the hyperscaler space that's kind of one of the core areas of opportunity because

36:16 you're putting in lots of cabling lots of of of um you know metal working and

36:23 materials and don't forget that has an energy usage that's further down the line but it's equally as important so

36:29 ensuring that you're looking at that whole process from you know the concrete in in in your network Center all the way

36:36 up through the tin the metal and every part of it you need to really be looking at all of that otherwise you might be

36:42 saving in one place only to be driving cost in another and I I don't see Telos

36:47 and service providers taking that infrastructure spend as strategic as they should and every year they have to

36:54 they have to spend this money because there's new health and safety laws or there's or there's things that just need replacing or there's new

37:01 environmental laws things like diesel container systems have to be continually updated and if you can update that and

37:07 build build a more Greener or more sustainable plan that can have a massive impact especially if you're working you

37:14 know and as merco said you know Colts organizations got thousands of sites all over the the globe um as you you're

37:21 going to be continually investing in them then think of that that core infrastructure spend much more

37:26 strategically than you do today okay thanks Neil end to endend thinking that's what we're talking about

37:33 here um now we're coming towards the uh end of our discussion here but we've got

37:39 one more question to to wrap up uh and that is for Network operators looking to

37:44 use any AI tools to improve their Network Energy Efficiency where should

37:50 they start Beth let's come to you first oh my God uh where should where

37:57 should they start well I think that the lwh hanging fruit is is um you know

38:03 focusing on um you know not necessarily AI uh focusing on you know uh automation

38:11 configuration as as Neil mentioned you know turning on the Energy Efficiency uh

38:17 configurations on your Hardware I and that's pretty simple right uh going through your network and and figuring

38:24 out what should be turned off um I had to laugh when I heard that story because

38:30 um years ago I I ran a lab and I had to move the lab and um you know I was

38:36 turning off all the equipment and there was this one box that didn't have a label on it and I didn't know what it

38:41 was and uh so you know my staff was like oh what should we do and I'm like well we have to turn it off well I found out

38:48 what it was it was the uh DNS uh one of the uh top level DNS servers uh for the

38:55 world um and uh so it quickly got turned back on again in a different location

39:01 with a label on it um so uh this you know just knowing what you have in your

39:08 network and and you know optimizing um you know the equipment to make sure that it does run as

39:14 efficiently as possible and of course uh obviously uh negotiating with the your

39:20 energy um your providers uh super important uh and none of these have

39:25 anything to do with AI but all good points so to come back to

39:32 Neil's point there Beth uh this endtoend thinking needs to include label management uh that's a lesson that uh to

39:40 to take away from your little anecdote there uh so Neil let's come to you um you know what uh where should Telos

39:48 start when they're thinking about using AI to improve their Network Energy Efficiency yeah I mean I mean I think

39:55 looking at the overall architecture Network how can I take layers out you know we've had this Optical layer in the network for a long time that I'm just

40:03 not convinced drives huge value and I think we can do a better job at at the routing layer and we see many Network

40:09 operators heading in that direction and actually it sounds really hard but actually it's relatively simple because

40:15 you because in most networks you've got some you know an element of redundancy and you can you can delayer quite simply

40:22 I think that's definitely one place to start I think the the and you know we've got in our platform we've got tool

40:27 called Paragon that can help you plan that I think the second thing I would say is is your own engineering expertise

40:35 um people are passionate about sustainability particularly engineers

40:41 and because they know that if we save some money on energy I can invest that

40:46 money on perhaps another service or some other capability in in you know in their

40:52 organization and I think there's so much data in in the kind of engineer head

40:57 that with a little bit of B little bit of extra bandwith they can make a big difference in sustainability and they

41:03 want to make that difference right um you know we see um a lot of uh our

41:10 customers being really proud about the fact that they have saved or they're kind of at the top of the league table

41:15 for having reduced energy year-over year and it and it really brings a a kind of added engagement benefit from your

41:22 engineering teams who want to be part of this and who've probably got most of the knowledge that that you need not all of

41:27 it but but quite a lot of it and then and then it's the last part I'll mention again into to end we have a a customer

41:34 here in the UK that's using Google Maps to pre-plan The Last Mile build of fiber

41:42 and that isn't a you know an electricity energy saving but it's it's an energy saving in that when they turn up to do

41:50 uh a job to build that last mile of fiber they get it right first time as opposed to turning up and realizing they

41:56 haven't got the right Vehicles they haven't got the right safety requirements or whatever it is um just

42:02 by thinking about your whole um you know your day-to-day process and and how that

42:08 affects energy usage both you know from the grid but also in in you know diesel

42:13 or Transportation costs that can also um be a big factor and and also as as a

42:19 vendor so you know Beth made made an interesting challenge about this not being top priority for vendors um I

42:26 think what's not priority top priority for vendors is telling about what a great job we're doing in this space and

42:32 and I'll give you one example the thickness of the metal in one of our devices can can be a differentiator

42:40 between how green a device is and how un ungreen it can be and we put a huge

42:47 amount of effort into every aspect from you know the the thickness of the metal

42:52 to how we mount it in cabinets to the packaging um to ensure all of that is as

42:58 as as sustainable as it as it can be and actually I think we've done a bad job of sharing that information with the world

43:05 because it's one of the things when I when I joined Juniper a couple of years ago one of the things I was most impressed about was just we had this um

43:13 great story but it wasn't really well understood for sure we've got more to do and we continue to push that but we want

43:20 and we want to work with our customers to see how we continue to drive those optimizations but that those are some of

43:26 the places that I would I would be starting with but and and the to reemphasize one of them your own

43:32 engineering team because they've got the knowledge in their heads for sure yeah well I mean you've always advocated

43:38 talking to customers talking to staff and there's obviously uh a lot of gain to be had from that and we could all use

43:45 some uh positive stories from the industry as well and use cases so we we would encourage uh all companies to to

43:52 share what they're doing to to make a difference uh merco let's come to you and then we'll end with Diego seems to

43:59 be the answer to start with is actually don't don't use AI uh the basic has been

44:05 said I think if you look at the efficiency you can get with uh

44:11 modernizing the network is is amazing we we been able to uh reduce energy

44:18 consumption of certain layers of the network at the at the at the routing layer at the I layer by over 90%

44:26 introducing coherent Optics and and uh refreshing the technology with the newer

44:31 generation technology which is heavily and energy efficient I don't think you

44:37 you really should embrace uh uh using AI unless you have done the basics because

44:43 you you're going to try and optimize on a on a non-efficient network having done

44:49 that I think you need to First Look at the use cases what are the things that could bring more more efficiency there

44:56 is no uh I think from our experience there is no tool you can buy today that can help you specifically optimize uh

45:04 energy consumption based based on AI so a lot of this uh what we have been doing

45:10 is actually internal build so you have to think about how you're going to how you going to create the capability to to

45:17 enable uh uh your company and this is actually a Gile development to to create

45:23 these use cases for uh for AI okay thanks MCO and uh Diego final word to

45:30 you uh where should operators start when they're thinking about um when how or if

45:37 to use AI to improve their Energy Efficiency whe would to what my colleages have said this to to use data

45:46 you have to you have to understand where your data sources is are you have to

45:52 understand where your data preprocessing uh an identification

45:58 should be located you have to generate massive amounts of data of data of of

46:03 many different kinds I'm not talking only about uh the raw data that you can

46:09 use for feeding an AI or or training the AI I'm talking as well about data's

46:15 knowledge these Golden Nug nuggets that Neil was mentioning at the beginning in

46:21 the brains of many many engineers useing at the seed of what you start to do and

46:27 then something that is important once you have in place those data something that is important as well is that you

46:33 take into account that normally you would need data unusual data to train

46:39 the systems to validate the the systems to be sure that uh that the system the a

46:45 system is going to behave approv in in in unusual cases and for this you need

46:52 mechanisms for either share the data with some uh some of your of your of

46:58 your fellow U participants in the in the Telco industry or generate them by means

47:05 of what we were mentioning before digital TNS or or synthetic environments

47:10 in which you can experiment but whatever the case AI is based on data ai ai

47:17 consumes and produce data massive amounts of data and we need to start

47:23 with H with this accurate and trustworthy data okay great insights there and and

47:29 clearly one of the key takeaways from today is that it's really important that Network operators mine not only the data

47:37 they have in their network but also the data and information and expertise that

47:42 they have in their Engineers uh but we must leave it there I'm sure we'll continue this debate during our live Q&A

47:49 show later and for now thank you all for taking part in our discussion

47:57 and if you're watching this on day one of the green Network Summit then please send us your questions and we'll answer

48:04 them in our live Q&A show which starts at 400 p.m. UK time I'm sure there are

48:10 many questions we still need to address so let us know what you want to hear

48:16 discussed now the full schedule of programs and speakers can be found on the Telecom TV website and that's where

48:22 you will also find the Q&A form and our poll question for this Summit for now

48:28 though thank you for watching and goodbye

48:34 [Music]

48:41 [Music]