The role of software in reducing IT infrastructure emissions

Behrad Babaee:

Okay. So hi everybody. My name is Behrad Babaee. I'm a technology evangelist in Aerospike. Aerospike is a NoSQL database company. My history with NoSQL databases is quite extensive. I've been in Aerospike for more than two years. Before Aerospike, I was in another NoSQL database company called DataStax. They are the main contributor to the Apache Cassandra project. Before that, I was mainly a developer. I've been using Cassandra heavily. Before that I used Couchbase for a short stint. I used MongoDB extensively back in the day. Well, let me put it this way, that the first time that I touched NoSQL is 2010 when I developed application on HBase. So I have a bit of a background on that. So that's a quick introduction about me and I'm joined by Abhijit. Abhijit, do you want to give an introduction to yourself?

Abhijit Sunil:

Absolutely. Well, thank you everyone for joining. It's a privilege to talk to all of you. I'm Abhijit Sunil, I'm a senior analyst with Forrester Research. I particularly work in sustainability and cover the IT decarbonization levers that various industries undertake at this time. I've been with Forrester for over four years now, before which I was a researcher at McKinsey. And I am looking forward to having a discussion with Behrad on some of the recent research that he did in software sustainability and also presenting some of the findings that we've had at Forrester on a survey that we did worldwide on the state of IT sustainability, the trends that we observed and how and where emissions in IT come from or within the enterprise architecture of all industries.

Behrad Babaee:

That's great. Thank you very much. So I want to start the conversation with setting the scene. So the title of this talk is the role of software in reducing CO2 emissions. Usually when I talk about this topic to people, the first question is that, but software is something virtual, right? Something virtual doesn't necessarily generate emissions and toxic waste. And for that, I have an analogy so I want to give that so we can start the conversation. We rarely talk about emissions of jet engines. We usually talk about emissions of aviation industry. We are talking about emissions of flying, right? Although it's actually the thing that is actually generating the emissions is the jet engines. So software industry and hardware industry are similar to that. It's the software that is running on the hardware and because of that you need the hardware. If you turn off all of your software systems, you wouldn't need any hardware.

And if you don't need any hardware, you can turn off all of your data centers and you're not going to produce any emissions. So it's the driving of emissions. Of course like aviation industry, efficiency of the jet engine is important as well. You cannot just forget about that. So if you want to have effective conversation, we have to focus on reducing the emissions on the hardware infrastructure side but we shouldn't forget about the efficiency of our software platforms. So I have done a bunch of research about software efficiency and effect of it on reducing carbon emissions and as Abhijit said, he led a very [inaudible 00:04:23] research or survey on the state of sustainability in IT organizations and hopefully putting our knowledge together, we are going to give you some information that you will find useful. So with that, I want to start the conversation. So Abhijit, in the beginning of your report you define basically the [inaudible 00:04:58] sources of CO2 emission organizations. Can you explain those basically sources so we can basically see the differences between them?

Abhijit Sunil:

Sure. So you set the stage very well, Behrad, in talking about how sometimes we don't think about software as having carbon emissions, but it is really the underlying infrastructure that powers the sources where carbon emissions of software comes from. But if we look at the IT organization within any company as a whole, where do emissions come from? So we found in our research that emissions came from a variety of sources within the IT stack. First and foremost, is the data center and cloud infrastructure of organizations. Now data centers themselves are a big consumer of energy globally. The International Energy Agency recently published a report that said that only data centers globally consumed more than 1.3 to 1.5% of global final energy demand, which is substantial and likely to grow even further. So in industries where data centers form a major backbone of their operations, for example financial services, data centers are a major source of their scope one and scope two emission... As well in just a moment.

But other than data centers, then are the emissions that come from end user devices of an organization's IT infrastructure that employees use at the workplace. So essentially it in the workplace and emissions associated with that. And then a third element, very important, is applications and the emissions that come from the software part of enterprise architecture. So the emissions that come from hosting applications to the emissions that occur during software development, all of that falls into that umbrella. And the sources from where carbon emissions come from can be then subdivided into the carbon emissions that occur when infrastructure is manufactured to the supply chain, then infrastructure shipped to premises. And then also at the end of life of infrastructure, the angle that we talk about in terms of circular economy and how we define those.

But we just touched upon something that is a very widely used framework, scope one, two, three emissions. And I'd like to take a moment to define what those are. In any organization, scope one emissions are direct emissions from the organization. So if a company burns something on its campus, if there are emissions coming out of factories for example, those are all classified in the scope one emissions and direct emissions of an organization or of any entity. And then scope two emissions are emissions that occur due to the energy that an organization consumes. So these are also directly controllable because an organization can choose to consume one form of energy or the other, clean versus conventional energy or consume less energy. So that too is directly under somebody's control, but it is occurring at the place where the energy is generated. So that's classified into scope two emissions.

Everything else including upstream and downstream supply chain, the emissions that occur due to employee commuting or business travel, all of that is classified under scope three emissions. And they are all set to be indirect emissions because it occurs in the supply chain or occurs because our suppliers emitted those carbon emissions in down or upstream supply chain. And it's important to look at this framework for scope one, scope two, scope three emissions because it gives us a good way to classify and measure the carbon emissions of any entity for that matter. This framework was developed by the GHG protocol and is widely used in all of the reporting frameworks that we currently have. And those include even the government regulations that include CSRD or the way in which the US SCC has been proposing how organizations should be measuring their carbon emissions and report them. So this is the definition of scope one, two, three emissions, which we'll talk about quite frequently now even in our conversation.

Behrad Babaee:

I think I was on mute. Sorry. Yeah, thank you very much for a very comprehensive explanation of the state of system. So one of the things that you touched on was the amount of energy that the software systems are using. And I was reading something interesting the other day that was about the amount of energy that producing semiconductors is actually using. We usually talk about the amount of energy that the data centers are using. And as you said, it's estimated to be something between one to 4% of the energy produced in the world. But there's something around one to 2% of the energy that is produced in the world is being used for generating just the semiconductors, right? So it's a massive number as well. It's estimated that the amount of energy that by 2030 is going to be used for generating semiconductors is going to be as much as the energy that a country as big as Australia is using today.

Or the amount of emissions given that we are moving towards greener energy, given that, the amount of carbon emissions that these semiconductor manufacturing is going to produce is going to be the same as the amount of carbon emissions as a country is big as Portugal is generating at the moment. But how do you see this trend? Do you think that it is going to continue? Do you think that it is plateauing? How do you see the trend of energy usage in general in the IT industry?

Abhijit Sunil:

It's a great point, Behrad. A lot of indicators in the market at this time lead us to think that the energy consumption of IT will grow, but there are more nuances to that. When we look at the consumption of energy by data centers over the past decade, it has been more or less flat despite the fact that data center workloads have increased exponentially. So we've seen more and more penetration in the telecommunication sector, more SaaS applications come into the market and the movement towards the SaaS model by even traditional software vendors, all of which indicated that there has been a steady rise in cloud and public cloud consumption. And we would think that there would be a proportionate increase in the consumption of energy by all of these data centers that host these applications. But that's not been the case. It has actually been somewhat flat and that's because of all of the innovations that has happened in the data center industry.

We've seen a lot of passive and active sustainability related optimization in the data centers. As an example, passive optimizations include actually physically geolocating a data center in an area where those data centers can rely on its environment for cooling purposes. For example, moving towards anatics or closer towards the poles where there's cooler atmosphere just available rather than relying on air conditioning for cooling these data centers to achieve a good power usage effectiveness internally. Then there are active measures that we have seen rise, for example, the use of artificial intelligence to monitor and anticipate the energy consumption data halls. And we've seen some research that has actually told us about substantial energy savings because of such measures. So because of all of this and the optimization that has continuously occurred in the networking space and in data transmission, we've seen that the power consumption in the data center space has not increased at least in proportion with the amount of workloads that has gone to the cloud or the energy or the demand in data center services.

However, a major trend that we will see now is the advent of specialized workloads including artificial intelligence. And we all know about generative AI and the trends that itself is creating in other parts of our workplace. So therefore, we'll see a lot more in applications that are very computing intensive. And at first [inaudible 00:15:26] we should report, which we titled Jekyll and Hyde, which looked at the both good and bad parts of emerging technologies including artificial intelligence and other emerging trends that we see in the market right now. And we looked at how some of these trends could aid in sustainability, as an example, the AI energy modeling that we can do within data halls that I just talked about. And on the other hand, the energy consumption that is caused because of AI. So how do we balance that?

This balance I think will be a major trend that we'll see going forward and more innovation that we will need in the space for tipping the balance towards being more efficient and more optimized will be something that we'll see more and more of and more investments will come forward for the space. And more discussions will occur around the balance between how AI or other emerging technologies will either aid or hinder sustainability activities in this space. So we'll generally see more demand for energy by IT as we actually will see more digital transformation efforts globally, but at the same time I think we'll see more innovation happening so that we can balance that curve as much as possible.

Behrad Babaee:

Yeah. Yeah, thank you very much. And in your survey you look at the state of reporting at the moment in the industry, so can you explain a little bit about that? How's this state of reporting? How companies are actually trying to report the emissions? What they are doing? What are the trends there and what do you see in the future of these activities?

Abhijit Sunil:

It's a great point. In fact, sustainability reporting we found in our survey is mainly driven due to compliance at this time. So if we looked at it as if it is a carrots and sticks approach, sticks seem to be driving a lot of sustainability reporting and we are seeing evidence of this globally when we see the advent of more compliance and regulatory pressure for organizations to report their scope one, two, three emissions and beyond, not just their carbon emissions but overall their ESG metrics. So one interesting development is CSRD in Europe where public companies need to report into CSRD, the various ESG metrics that they are now mandated to measure and report against. And we are also seeing that some of the challenges in the space are in rightfully measuring scope three emissions. So we talked about how scope one and two are direct emissions of an organization and scope three emissions are indirect emissions.

So for an organization in the financial services space, for example, data centers form the majority of their scope one and scope two emissions. This also means that it may be easier to actually measure those and report it than scope three emissions, which should involve collecting data from multiple suppliers penetrating deep into your supply chain. And this has been a major challenge across industry and we found that in our survey as well that scope one two emissions and related double clicks into those, for example, air and water quality monitoring, those were aspects metrics that were measured and reported much more than scope three emissions. And then in terms of frameworks, we found that government mandates and reporting compliance related activity around the globe is actually driving more reporting and not just in the public sector but private sector actually needing to catch up with their counterparts in the public sector simply because then this leads to more visibility amongst clients.

And we are seeing this as an increasing trend, how organizations are considering sustainability reporting as a differentiating factor within the industry as well. So the government related compliance issues and regulations actually drive a lot of the first steps in terms of how an organization should measure their scope one, two, three emissions and set internal processes for measuring and reporting. But at the same time then using that as a baseline, organizations are setting targets and really sending the message out to their customers and clients that they're doing their part to be responsible for the environment. So we are seeing an increased trend where the models of sustainability will continue to grow.

Behrad Babaee:

Yeah, makes sense. And I also think there is another problem with reporting scope three emissions is that scope three emissions of this company is the scope one or two of another company. So it's very easy to double report them or triple report them, if you will. So yeah, scope three is very difficult, but reporting it is still something that companies should try doing because you can always just offload everything that you do to someone else and just forget about that. There are some complications in that, but I think it is something that we still need to aspire to do, which I think you agree that most companies at the moment are not doing, right?

Abhijit Sunil:

That's right. And most reporting frameworks also mandated the research and collection of metrics around scope one and two emissions, but not quite scope three emissions so far, but we are seeing more and more of that come forward. And there is also the problem of lacking standards in the space, exactly like you pointed out there, somebody's scope one or two emissions becomes the scope three emissions down the supply chain. So for example, if we are talking about an end user device that an organization uses, the manufacturer of that device or the embodied carbon that's associated with the manufacturer of that device is the scope one or two emissions of the manufacturer or the vendor who manufactures that device. But it becomes a scope three emissions as it comes through the supply chain and is being used in the organization of the client who purchases that device and starts using it.

So these are all aspects that are very hard to measure so far, but at the same time there is more research and best practices that are being discussed and we'll hopefully see more organizations measure and report these types of metrics as well, which will be very crucial for us to then think about reduction in government emissions across the value chain.

Behrad Babaee:

Yeah, that's very good. So one thing that I wanted to talk about was the efficiency of software and the effective efficiency of software and reducing the emissions. And for that I want to basically look a little bit back in the history of computing because that's helpful. If you look at from the beginning of the computing, so around 2005, we spend everything that we have on making more efficient software, right? Because efficiency is the most important thing. Everybody's trying to make things more efficient because firstly, our CPUs were not fast enough. So if you could make something more efficient, it was running faster. Also, we didn't really have a scalability. If you could do something faster, your CPU could do more things. But around 2005 with advent of internet and internet scale companies, that efficiency wasn't that useful anymore, right? So we started to care less about efficiency and basically the talk of the market has started to become a scalability.

So we started focusing on a scalability instead of efficiency. And it was right because it didn't matter how efficient your software was. Without a scalability, you couldn't make Google, you couldn't make Yahoo, you couldn't make any of these big internet companies that we know today. So we started focusing on the scalability and we forgot about efficiency, but one thing that is important is that yes, a scalability is a great thing because if you have a workload today that you can handle it with this software, if for whatever reason six months later this workload is doubled, you can just double the amount of hardware that you have and then this scalable platform that you have is going to handle that workload. So that's great, right? So your business would never stop. But there is something that nobody usually think about and is that if you have a software that is 50% more efficient, right? You can do that same thing that the double amount of workload with the same amount of hardware. So you don't need to increase the amount of hardware that you require if you choose to use more efficient software, right?

Then the next question become this that okay, yeah, I agree that more efficient software can be that much more efficient, but do we really have software that is 50% more efficient? And the answer to that is yes, the algorithms are not changed, right? Most of the algorithms that we use today are still the same algorithms that we designed in possibly '70s and '80s, right? There's nothing new. But there's been a seismic shift in the amount of hardware that we have in different parts of the computer other than the CPU. For example, the amount of RAM that you get on a modern machine can be up to 24 terabytes on a single server, right? Back in the day, if you go 10 years ago, a little bit more, 18 years ago, the maximum amount of RAM you could have on a single machine with 64 gigabytes, right? That's around 350 times difference. And the speed of RAM since then has increased around 10 times as well. So you can utilize the modern hardware in a different way to produce more efficient software, right?

And Well, Aerospike is basically using these changes in the hardware industry to produce a more efficient software. And I did a research last year which I wanted to discuss the result of it in this talk as well. So as I said at the beginning of the talk, I used to work for a company that's producing most of the code in Apache Cassandra and I used Apache Cassandra for a number of years before joining that company as well. So I know Apache Cassandra, if not better than Aerospike, at least as good as Aerospike. So one thing that I noticed as soon as I joined Aerospike was that this system is much more efficient than Cassandra. And I decided to basically do a comparison between the efficiency of these two systems and what's better than comparing the efficiency in terms of CO2 emissions.

So the paradigm is relatively simple. I chose a specific amount of data that I want to store on Cassandra and that same specific amount on Aerospike. Then I said, "Okay, how much resources do I need to handle this much data on Aerospike and how much resources I need to handle this much data on Cassandra?" And then I calculated the amount of emissions that these resources generate, both in terms of the energy usage and the amount of energy that was used for producing these hardware resources. And the final result was around 80%. So you can reduce the amount of emissions by moving from a less efficient software to a more efficient software, but 80%. The silver lining here is that you're basically reducing the size of your infrastructure by 80%. That's why you're reducing your emissions by 80%, right? It's just directly. Even if I didn't do the math to calculate the exact tonnage of the CO2 emissions, you should be able to say that because the amount of resources is reduced by 80%.

So when you reduce the amount of your resources, it becomes cheaper as well, right? It's not only you are reducing your CO2 because now your infrastructure footprint is smaller, you're going to spend less money and then efficiency also means fasting. So you expect that everything be 80% faster as well, which, I mean, is a pretty good thing. So yeah, there are many different things, many different reasons that people should look at using more efficient software as one of the ways that they can reduce their CO2 emissions because 80%... I don't really think that there are an easy way that you can reduce 80% of CO2 emissions of a data center. We are talking about smaller margins there because we thought about that a lot and we tried to optimize that a lot. This side is something that we haven't thought about and I think that as we are moving forward, as the size of the data doubles every now and then and we need to basically increase the amount of hardware that we need, that's the type of innovation that we need to think about.

Abhijit Sunil:

You're absolutely right. Better than pointing out that as the amount of data the world increases, we have to be thinking about scaling and efficiency in software. And we talked about earlier and how in the IT stack the emissions come from three major areas. One is IT data centers and cloud, and the other is IT in the workplace, and then the third part is about software and applications development. Then all of this work that you've been talking about, your research covers that software part and how to make it more efficient. And there are several debates here like you rightly pointed out. For example, fast is faster, more efficient versus optimized and all of those types of debates on how do we reduce the amount of man-hours in software development, will that lead to more efficient software or will that lead to less carbon emissions throughout?

And these are really interesting frameworks to think about. One of the things that I had had come across and had been working closely with some clients on is to understand the right refresh cycles of infrastructure. So what constitutes that decision? Now when we think about whether we should be upgrading any infrastructure to take away, for an example, let's say a server within the data center. Then how do we make that decision on whether upgrading that server today versus one year from now will give us any benefits in terms of carbon emissions? So things that we should consider are about how efficient the new server would be, but at the same time, what would be the embodied carbon that comes with manufacturing of a new server? Then what about disposal of our older legacy infrastructure? What would be the carbon and ecological effects of that? And then these supply chain of bringing the manufactured new efficient server to our premises and taking the legacy infrastructure off the premises.

So all of those things combined makes the argument for whether replacing or upgrading infrastructure today versus amortizing the carbon footprint of that infrastructure for one more year will give us any more or less benefit. So it is very interesting and I'd love to hear more about any more real world examples that you have as well, better that backs your research.

Behrad Babaee:

Yeah, that's a good point. Actually we do have two publicly available use cases that I can talk about. So one of our customers is a company called Critel. They are one of the biggest ag tech companies in the world. They're the largest in Europe. They migrated from some other database technologies to Aerospike three years ago if I'm right. Yeah, it's three years ago. And in the report that they published, they mentioned that one of the reasons that they did that was basically reducing their CO2 emissions, and in that report they mentioned around 80%, which again, very similar to the number that I produced. But we have something that is even newer than that and that one was even more surprising. So another customer that we acquired in the last few months actually is the famous company, TomTom, the navigation system that you possibly still remember. They're still around and they're doing lots of things in the background. You don't have the devices anymore, but possibly your cars are still connected to their systems in the background. They are doing mappings and navigation for lots of the largest car manufacturing companies.

So they were using another caching technology actually. And they decided that instead of using that caching technology, they want to use Aerospike, which is a database. But this speed of aerospike is very similar to caching technologies. As I said, it's so fast that instead of using cache, you can just use the database and it will be as fast as the cache. And well, of course it uses much fewer resources and it's going to end up being a lot cheaper because, well, keeping everything in RAM is very expensive in comparison to keeping everything on disk. So they moved from the caching technology into Aerospike and they've been on the technology for five, six months. So they reported lots of positive things. So for example, they said that the number of outages that they had was reduced from, I don't know, six outages every two weeks to zero in the past six months. So that's already good.

But one thing that is relevant to this topic is that they reduce the amount of their CO2 emissions by 86%. Well, it is important because firstly, it says that companies actually do care. In the public announcement of moving to this technology, they mentioned the amount of CO2 emissions that they could reduce, and we didn't ask them to do that. So they mentioned that one of the reasons that they moved is reducing the emissions and they calculated the amount of emissions that they could reduce. And again, it was basically by reducing the number of surveys that they required, they used to have, if I remember correctly, 49 surveys or 48 surveys and they reduced it to three basically.

Abhijit Sunil:

Oh, very interesting. That leads me to think-

Behrad Babaee:

Topic that I said about reduce the amount of resources. Go ahead.

Abhijit Sunil:

Oh, sorry to interrupt. Please go on.

Behrad Babaee:

Yeah, so I was just saying that they could basically reduce the amount of emissions by that. So it's been very, very interesting. So as we are closing to the end of the hour, well, the 45 minutes that we want to have this, I want to ask you a final question. So how do you think companies can create sustainability strategy for themselves so they can reduce their emissions planet and get there and be able to report good numbers? Because now as you said, a lot of legislations are asking companies to report that, and I think nobody wants to be in the bad name, right? Everybody wants to report good numbers. So how companies can reduce their CO2 emissions moving forward?

Abhijit Sunil:

Well, developing a good IT sustainability strategy really begins with understanding the IT estate of your organization and looking at where emissions come from and then creating a map of what are some of the most important and impactful decarbonization levers versus what are most pragmatic. As an example, we talked about how scope two emissions is about energy consumption. Data centers consume a lot of energy, but it may not be possible for many organizations to really switch to green energy in many regions that they operate in. On the other hand are strategies associated with end user devices and circular economy that may or may not be possible or may have more or less impact compared to some of the other investments that they could make in the IT sustainability and decarbonization efforts that they would make. So in our research we found that the most successful strategies involve creating a framework for thought and then creating an analysis of the most impactful decarbonization levers versus the most pragmatic.

And when we think about the decarbonization levers, in fact it's vivid from the examples that you gave also, Behrad, about the thought process of looking at where emissions come from. So hardware and data centers is one big bucket. Then like we mentioned, the end user devices in IT in the workplace is another. And the third aspect is about software development and application sustainability. And one of the trends that we are seeing now is about how difficult it is to actually measure the carbon footprint of applications itself, but there are more processes and standardization that are happening. And for example, we are seeing more and more best practices in software development as well as using proxies such as measuring the optimization and cloud workloads to understand how much data or how much carbon emissions actually come from cloud workloads itself. So those are some of the key aspects of creating an IT sustainability strategy.

First to look at an end-to-end enterprise architecture decarbonization strategy in terms of where emissions come from and grouped into these different buckets that we talked about. And then creating a list of some of the most impactful decarbonization levers versus the most pragmatic, and then creating more processes associated with that in terms of what types of metrics and KPIs can you measure and how do you then optimize on those KPIs and metrics. I'm curious to hear your thoughts as well, Behrad, on how the software industry can itself enable more decision makers to make better decisions regarding IT emissions.

Behrad Babaee:

Well, I think the first step in doing any engineering work is having a framework to measure the impact of decisions that you make. I mean, it's a very common thing for a software designer to look at some non-functional requirements when they are choosing different technologies that they want to use. So non-functional requirements are, I don't know, for example, the SLA, how fast you want this system to be? Right? Or I don't know, what is the cost of this system should end up? These are not necessarily defining the functionality of the system, they're talking about the expectation that you will have from the system. So I think one of the non-functional requirements that people should pay attention to is the amount of CO2 emissions that the software platform is going to generate.

To have that, you have to have basically a framework to basically look at and say that, okay, if I use that technology, it's going to have these, I don't know, outputs and it's going to generate this much CO2 emission, and if I use the other one, it's going to have these characteristics and that much CO2 emissions. And that should become part of the conversation when you are choosing which technology you want to use. And for this, we need some standardization. I don't know, very similar to when you buy a refrigerator, for example, it has a sticker that says that it uses, I don't know, energy level A or B or C. Software can have something like that as well. I don't know, I'm just thinking out of the box that different technologies could get standards for amount of resources that they [inaudible 00:42:11].

I think that could be a step forward that allow decision makers to make more informed decisions because at the moment, nobody talks about this. You just say that I want a, I don't know, a streaming technology and you just pick the streaming technology that is the best for your budget and fulfill your other requirements, not the requirements related to [inaudible 00:42:38] emissions. So the conversation is very interesting and I wish we could continue, but we are already at the top of the 45 minutes. Thanks to everybody who joined and listened. And if there's nothing more, I want to say goodbye. And Abhijit, if you have anything else to say, please go ahead.

Abhijit Sunil:

Thanks everyone for joining and this was a very interesting discussion. And Behrad's research on software sustainability is really crucial because of the way in which we should be thinking more about standardization and measuring how software can be made more sustainable, and looking into the ways in which software can be made more efficient in itself and how the software industry can overall become more sustainable. So those are really interesting outcomes that we discussed and was great to hear from you, Behrad. So thank you very much everyone for joining and hopefully there were some great takeaways and always happy to answer more questions too.

Behrad Babaee:

Thanks everybody and good to see you. Goodbye.