Interview with Ender from DSTECH
Ender Demirel, COO and Co-Founder of DSTECH, talks about the progress of the ECO-Qube project, what attracted them to participate, project outcomes, challenges, and achievements to date.
As part of the ECO-Qube project, DSTECH focuses on CFD simulations, including researching temporal variations of flow and thermal structures inside the data center.
You can watch the whole interview here, or you can read the transcript below.
Transcript of the interview
Good afternoon. So for this series, it's now the third interview in our series and I'll be interviewing Ender Demirel. Ender works at DSTech - Design & Simulation Technologies. It's an SME that was started in 2019 and he's focusing on the CFD element of the ECO-Qube project. Ender welcome! Maybe you can give us a little bit of background about yourself and about DSTech.
Thank you, Mohan. My name is Ender Demirel. I'm the co-founder of DSTech and I'm leading the ECO-Qube project at DSTech. The main role of DSTech at the ECO-Qube project is to perform some numerical simulations in order to understand the flow and thermal structure inside pilot data centers located in different regions of Europe. So, we are currently working on CFD simulations and then we will carry out some experimental measurements in order to calibrate and validate our numerical models.
Perfect! So what attracted you to the ECO-Qube project in the first place?
Well, research studies in the literature generally focus on the increasing energy efficiency of data centers. However, the establishment of a holistic approach based on real time sensor data and CFD simulation results for different thermal scenarios attracted me to the ECO-Qube project. I believe that the results of the ECO-Qube project will lead to the existing technology in data centers in the near future.
Right. So it's the holistic approach and specifically applied to small data centers where we see growth and where we foresee a large growth. What is innovative about the ECO-Qube project?
Flow and thermal structure may significantly change inside the data center. So the cooling system should be sensitive to this fundamental feature in order to reduce energy consumption in data centers. The development of a zonal heat management approach is one of the innovative parts of the ECO-Qube project. In order to account for special and temporal variations of thermal structures such as recirculation hot points and short circuit effects that may reduce cooling efficiency of the system. The second innovative part of the ECO-Qube project is to develop an artificial intelligence for the smart CPU workload orchestration. In addition to room type data centers, we will conduct research and development studies on air type micro data centers in one of the pilot sites of the ECO-Qube project. Applications of micro data centers in buildings will increase the new feature of edge computing in my opinion. Thus, research and development studies on a completely air type micro data center will be another part of the ECO-Qube project.
So in layman's terms, the heat distribution within a data center, the air, it’s different in different areas of the data center. So the idea is we start to model that to make existing IT efficient. And then the second part of what you said is to develop an AI, which can then optimize for it. So it's not just observing it, it's actually being able to react to it or even predict it. And that we think will give us far more efficiency in data centers. Is that right?
Yes, it's right. Consider that the cooling system usually detects the flow and thermal structure inside the data center. So the performance of the cooling system should be sensitive to the dynamic structure of data centers.
Interesting. And what are the main objectives and goals of your work package?
As the DSTech team, we are performing CFD simulations for different thermal scenarios in pilot data centers located in different regions of Europe. We need to know special and temporal variations of flow and thermal structures inside the data center. In order to address this need, CFD simulations will be carried out for the extreme conditions to review the thermal structure of the corresponding data centers. Moreover, experimental measurements will be carried out at pilot data centers to validate and further collaborate our numerical model. Specifications of the required measurement system, like measurement range, sampling frequency, and critical measurement points, will be determined from CFD simulations.
So how similar are CFDs from digital twins? We often hear about digital twins. And actually, in order to conduct CFDs, you need to have a model of the data center in the first place. So you're almost doing a digital twin and then modeling it. Right?
Yes, you are right.
Interesting. What are the biggest challenges facing you in this project?
The biggest challenge that we faced, especially at the initial stage of the ECO-Qube project, was to obtain the data required for the development of the CFD model for the pilot data centers. The participation of Bitnet Data Center to the project consortium as a partner, increased the momentum of project studies due to their responsible approach during the project studies.
Interesting, so one of the tricky things, especially what we talked about earlier, part 2, which is creating artificial intelligence, is that you need a huge amount of data and getting this data out of the data centers is difficult, not least because every pilot data center is different. Right?
Yes, they are completely different. We need geometrical data, the layout of the servers or cabinets, locations of the cabinets, information about the cooling system, and so on. We need to have some details about the data center in order to create a numerical model and to perform some numerical simulations. Sometimes it may be difficult to provide this data to the CFD simulation engineer, but the partners in the ECO-Qube project are trying to provide this data for the CFD simulations. I want to thank them.
And that's what's really quite novel, because this data never normally comes together. We have the silos in the industry, potentially for good reason, but traditionally they never shared this data. They've never interconnected the building energy system, the sort of understanding of the energy that's coming in from the grid and how green that is, the IT and the geometrical elements of the data center. Like what you said: That's a complexity and a challenge for this project.
Yes. Another outcome of the ECO-Qube project will be what type of data needs to be provided in order to perform some numerical simulations. The companies will prepare this data for CFD engineers and they can immediately provide this data for the CFD simulations.
So that is actually one potential output of this project is that people will understand if we can deliver XYZ data, then we can actually apply this AI and make our data center smart. So almost actually adjusting how we look at data centers and maybe adjusting how they're designed and built or at least where we put the sensors they might come as standard in future.
Yes, you are right. We are so lucky in the ECO-Qube project because the partners, especially the pilot data centers, are so responsive during the project studies. They are trying to provide this data in order to move forward in the project studies.
Awesome. What are the revolutionary effects? What impact could the development of the ECO-Qube project have on the rest of the industry?
We know that the operation of data centers is challenging for engineers due to their inherent dynamic structure. The ECO-Qube project aims at providing several solutions for data centers. Development of an artificial intelligence augmented decision making system based on a zonal heat management approach will be the most revolutionary effect of this project on the data center world. In the ECO-Qube project we intend to establish a holistic approach for the data center hardware and software systems to achieve the best energy efficiency performance of data centers.
And what is the progress of the project so far? Because what Jurg was saying last week was that we're about a third of the way in and we're about to connect the data centers to the data collection platforms that are about to start receiving data. But for you specifically, how does that affect you?
Well, DSTech is the leader of work package 2 in the ECO-Qube project. In the scope of work package 2 we are developing a new open-source software data center DST, which specialises in the CFD simulations of data centers based on open-source libraries for open-box and black-box modeling of server components. In our numerical model several novel boundary conditions were developed and incorporated into the data center DST software. Another significant progress in our work package is the calculation of efficiency indexes at each server and cooling unit inside the data center, which is fundamental to the development of the zonal heat management approach in the ECO-Qube project. This is very important for us and a very important progress in our work package. Finally I see the overall process in the ECO-Qube project as wholly positive.
Amazing. So what you are doing is going far more granular than anyone has gone before in a data center? Is that right?
Going right down to the server level and sort of really making that data center - basically putting a lot of sensors and I think you said metrics, or indexes?
Yes efficiency indexes or metrics. We are performing server level modeling in our numerical models, so instead of calculating efficiency or overall efficiency indexes inside the data center, we are calculating the efficiency indexes at each server and cooling unit. So it’s possible to detect which servers are consuming high energy, so we can detect the special variation of efficiency inside a data center. This is very important for the development of the zonal heat management approach in the ECO-Qube project. This is a different or novel idea for data center modeling in the literature.
A last question Ender: is there anything else you like to share about this project?
11 partners are available in the project consortium from different sectors and sometimes it may be difficult to create a common language between the partners. I want to thank the project coordinator Çağatay for his responsive and unifying approach between the work packages. And also I want to thank SDIA for the responsive approach for the technical alignments during the project studies.
I can attest that, we don’t talk from a common dictionary, put it that way, so that was one of the main challenges at the start. Fantastic Ender, thank you for joining us and thanks for your update on CFD in the ECO-Qube project. Brilliant, that’s all we have time for today. Next interview will be with Ali Serdar from one of the pilot data centers. He will be talking more about integrating a lot of what Jurg and Ender have said with the actual data center. Thank you.