The Critical Lowdown Podcast Episode 2
How can Open Networking meet the Challenge of the Climate Crisis? | John Laban - Open Compute Project
Demand for cloud services is growing and with that is the need for more infrastructure to support this growth in traffic. For every blog post, photo and tweet there is a physical manifestation somewhere in the world. These manifestations are Data Centers. The carbon footprint of Data Centers and their energy usage has come into sharp focus, as climate targets loom, and the climate crisis intensifies. How can open networking meet this challenge and bridge this gap?
Senior Systems Engineer for EPS Global, Barry McGinley, is joined by Open Compute Project board member John Laban to talk about how the switch to open networking can reduce emissions and energy usage in Data Centers.
Subscribe to The Critical Lowdown from EPS Global wherever you get your podcasts:
Transcript of Episode 2: How can Open Networking meet the Challenge of the Climate Crisis?
Ciara McCarthy: I'm joined today by Barry McGinley, Senior Systems Engineer for EMEA at EPS Global.
Barry, you recently had a chance to chat with John Laban at the Data Centers Ireland conference.
Barry McGinley: Hi Ciara, thanks for the intro. I was on the panel with John Laban at Data Centres Ireland which was in the weeks following the COP26 Summit in Glasgow, where Data Centers and carbon emissions were discussed at length, and how we can tackle this problem collaboratively, in particular with the Open Networking community.
John is involved with the Open Compute Project where he's concerned with decarbonization through open-source collaboration and also with the circular economy. He's also a board member of Open UK, and if there's a conversation about open-source or carbon emissions, John Laban is usually to be found there! So I began by asking him how Open Networking might help tackle this problem.
John Laban: When we start looking at the decarbonization agenda, we start to look at the idea of dematerializing the amount of hardware. If you start looking at network infrastructures, as you move away from the big block core switch concept towards a disaggregated solution, what you tend to find is you use less equipment and you use less connectivity as well, so if you're moving into a cloud-native type solution, data center for example, most of the copper cabling links disappear.
There's a huge amount of embodied carbon in the cabling links, copper, so we’re moving into 100G fiber links and it's a much simpler configuration at that level. It's not just the embodied elements and the energy elements in the difference between a mono-block switch, core switch, and the distributed array of spine switches and edge switches, it's also all the connectivity that goes between them that gets massively reduced. There's a big, big push when you go into enterprise Data Centers, lots of them are constrained by power. They're looking at how can we introduce more networking power, more server power, more storage capability, with the constraint that we've only got 1.2MW coming into this facility so that's helping them drive towards these solutions because they've seen the case studies, for example the London Internet Exchange (LINX) case study where the guide describes these savings that can unlock potential within the facility to put in more capacity for networking in a power-constrained data center. That's one big driver that people are going for.
Barry McGinley: So that will be to use less, and to use it better?
John Laban: Yes, and to get more for less as well.
Barry McGinley: We were on a panel earlier (Real Life Experiences: Reducing Energy Costs and Carbon Emissions in Data Centres using Open Source Hardware) and Niall (Niall Kelly, Director at The Cube – National Centre of Excellence for Low Carbon) said that only between 6 and 12% of a server is used yet it requires the same amount of electricity to power, and I was saying you could increase the usage of servers, not requiring any more power, through open software.
Moving onto virtualization. There’s VMs that everyone's afraid to touch, because you don't know what's to disappear. But managing from an open software perspective, being able to manage the VMs better, allows you to use less hardware and run a better data center basically.
John Laban: It does. Cloud is really about the virtualization of the facility, but the old network topologies in the enterprise data center are based on real instances, on real physical servers. Very little virtualization. With cloud everything's virtualized. Once you do that you get a lot more complexity in the network. If you look at network traffic in a cloud-based data center compared to a traditional enterprise data center, in the older traditional enterprise data you might have seen 20-30% of all the traffic flows are going between machines within the room, and about 70% was leaving the facility.
Barry McGinley: That was the 3-tiered hierarchical model they used to use?
John Laban: Yes, the old TIA hierarchy where you had the core switch, the intermediate, the solution. That has all changed. The profile of the pattern of traffic moving around there is all East-West, it's not leaving the facility. So today you might find 80% of all the traffic is between machines and only 20% is leaving the facility.
Barry McGinley: To build a network right using Spine-Leaf technology, you know how many hops it's going to take to get from server to server, which helps as well.
John Laban: In the old system, if you take the old hierarchy we've got a core switch, an intermediate, and then the top-of-rack switch, if you start counting the hops…
Barry McGinley: You don't actually know how many hops there are sometimes!
John Laban: You start to reach this five-hop problem and you start to bleed latency, whereas when you start to unbundle and just get this 2-Layer hierarchy it's very deterministic, you know what your latency is, and you can build in a lot more resilience into the network. I would argue you could go into a cloud-based data center using open-source type technology switching, with the clone network fabric, and you go in with a chopper and start chopping things up and it still works.
You can't do that in a traditional enterprise data center.
Barry McGinley: No, you take out any layer and you’re in serious trouble! I mentioned earlier you're on the board of OCP and Open UK, how much effect do these groups actually have? We work with the OCP ourselves, with the ONF, the Telecom Infra Project on the Telco side, what are they giving to the ecosystem?
John Laban: Here’s what these not-for-profits are trying to do. The Open Compute Project Foundation (OCP) started in 2011. The very first annual conference they had approximately 200 people in the room. Last year when we did an event we had 12,000 people. The biggest event conferences in the world are all to do with open-source technologies. Go to a Kubernetes event, there'll be 30,000 people at it. Just to give you an idea of what's happening, I became a board member of Open UK back in January 2020, we kicked it all off, we got 13 new board members in to integrate what we call the “3 Opens”: Open-source hardware, Open-source software, and Open data, because we know that if we can get these working synergistically, we know there's huge benefits financially, for agility, and the environmental benefits that come from it. I’ll give you an example of what's happening within these communities that are pulling together the diverse groups so they can integrate these complex systems. Open UK is only 2 years old, last week we held a conference at COP26. We had a venue which was a village, we had hundreds of people at it. When the Chief Executive, Amanda Brock, sent out a tweet, she hit a quarter of a million. I’ve been getting up to 1,000 new followers and connections every single month on LinkedIn alone!
Barry McGinley: Is this because of the carbon issues?
John Laban: Different drivers. One is people are really starting to say A) “I don't want vendor lock in”, B) “I don't want to pay all this money” and C) the environmental driver. The big, big focus in data center now is the fear of policy makers coming in, and creating moratoriums and closing down these facilities because they haven't thought enough about the carbon emissions and it's happened already, it's happened in Amsterdam, Singapore…
Barry McGinley: It’s coming to a town near you!
John Laban: It may come to a town near me now! We need to address the carbon emissions issue in Data Centers, there’s a big fear factor. It’s helping people leap into the open source direction. What these not-for-profit open-source communities do is the people within the group work together, and with other open source groups. So you mentioned ONF, TIP, there's a real synergy that's going on between OCP TIP, and ONF.
Barry McGinley: I’m surprised that there wasn't as much synergy before between the ONF and TIP, but with the ONF and the OCP, you can see it all the time, both groups are mentioning the other side because they're nothing without each other. They're trying to implement software solutions, for example Stratum and their CORD project, but it’s nothing without the hardware that comes from the OCP. So they should be close and they are, and it makes sense. Just on one point you mentioned – disaggregation. We have found that this is a big driver – vendor lock-in versus disaggregation, being able to be the master of your own destiny. How big a part do you think that has to play?
John Laban: Aggregation takes you out of the single monolithic vendor lock-in concept. I’ll give you an example of the benefits of disaggregation in the telco space. In the UK we had a problem. We had 3 vendors, with proprietary mono-block systems for telco infrastructures - Huawei, Ericsson, and Nokia. Then the UK government decided they shouldn't use Huawei. So they pushed them to one side and realized “wow”, for our critical telecommunications infrastructure, we've only got two vendors, and one of them was a bit dodgy in terms of where it was going to be in the future, so you didn't want to be left with just one vendor! The new strategy for the critical telco infrastructures in the UK is driving down an open-source route because the open route disaggregates the monolithic system and it allows more supply. Now we're in a position where instead of having 2 vendors, with this new open-source infrastructure strategy, we're at 20. We're going to be at 50 soon, and within a few years it will be 100.
Barry McGinley: We're working with a lot of Alt-Nets in the UK, they're springing up all over the place, and they're trying to introduce innovation, do things a little differently. I’m working with a lot of guys on projects for the Central Office, aggregating their OLTs, bare-metal can be used for just about anything in that area, so we have bare-metal OLTs, switches, routers, vBNGs, as you mentioned with the data center it's moved to virtualizing everything, and you've mentioned the RAN, Open RAN, virtualized in the distributed unit, the centralized unit, and so on. Very exciting times on the telecom side for us anyway.
John Laban: If I can bring up that one point that I made earlier about Rakuten, most people listening to this won't even know who they are, but they're a Japanese company that started doing online commerce like an equivalent of an Amazon, dominant in Japan. They saw that the mobile phone technology was crucial, so they are now a mobile phone operator. Initially they built out the traditional Japanese 5G infrastructure, using initially traditional solutions with the big monolithic players. But the CEO, Tariq Amin has been inspired by what he's seen in the Hyperscale Data Centers with open-source solutions! He's driven the new infrastructure towards fully disaggregated open-source solutions. Just to put that into perspective, they will finish building their 5G infrastructure in the whole of Japan 5 years ahead of schedule. They have dematerialized all of the compute function and lots of their Data Centers because they realized they don't need it anymore, because when you unbundle it, it doesn’t have to be in one position near the base station as a kind of small Data Center, you can take bits out and you can put it somewhere else. Your backhaul, fronthaul, whatever you want. It's been so successful they've now set up a new company called Rakuten Symphony, which is now a provider of telco infrastructure services to the world's telcos and the prediction is that by the end of the decade it will be a $100bn business.
Now that's how radical what's happening in the telco space is! Just watch on Open RAN, that's a magic of disaggregation. Put your stuff where you want to put it!
Barry McGinley: I have a little story about this as well, Cumulus was based on open-source software, it was the main player on the on the software side, the NOS side (Network Operating System) for bare-metal within enterprise Data Centers. They were a big success, but their customers who started moving into Campus Networks, normal enterprise networks, said to them “hold on a second, my data center is running on Cumulus Linux, I use specific tools maybe Ansible, Chef, Puppet, to automate that, can I not do that for my Campus Network as well?!” So that's how Cumulus got into the Campus Network side of things. EPS Global is a value-add distributor…
John Laban: You’re not just distributors anymore. When I first started working with you sold widgets, today I see you as: you're into software, you're into the integration business, and you're into the business where you truly make money. When you shift widgets, you might get high volume, but you get small margins, but now you're moving into sensible business I think which is the service business, and that's why you're there, so all good stuff!
Barry McGinley: Yes we started out selling optics (optical transceivers) and asked ourselves, what does that optic go into? - A switch, so we started working with Edgecore Networks. There was a gap between the value-added distributor and the end customer (because you end up dealing with a lot of end customers) that we had to fill. Because Open Networking was a new thing, it was a disruptor, it wanted to go in and take away from Cisco, Juniper and the incumbents, so it wasn't going to be easy. EPS Global had to do a lot of evangelizing, like the OCP does, take it a step further and make those relationships with the software vendors and take the customer on the whole journey.
But how can we do this better? There are integrators, there are resellers in the middle, not enough, and the distributors. We take the customer on the full journey, from looking at a piece of hardware, choosing the software that best suits the needs of their business, through network design, and implementation in the Data Center or in central office. How do we take more of the market share from Cisco, Juniper and Arista, and how do we tell people about this?
John Laban: If you look at the strategies of say Cisco, Cisco are rushing very quickly away from hardware. They're still maintaining sales, but how are they maintaining sales? They're buying software companies and their revenue streams. They are becoming a software services business. They are getting involved with OCP on this new project, because they bought a chip manufacturer recently and they've got these new chips, but they're not really into the switch infrastructures, their hardware is the silicon.
I think the realization amongst the big boys is they're not going to make money when you've got open networking technologies coming in, and they won't be able to compete anymore, because as the end users get more aware and they see these alternatives they'll think “hold on a minute, why am I paying all this money?” You've got the new way that they sell gear. You used to buy the whole package with the software from Cisco, but now you have to buy software subscription, so what it does is lowers the CAPEX, but if you want another 50 users, well tick this box and pay much more.
Barry McGinley: It starts to add up quickly.
John Laban: I think the real key thing is to bring awareness of open networking to the end users. I would say the transformation is going to occur one funeral at a time. Once new blood comes into an organization they will want to innovate, and that will make the change.
Barry McGinley: So unfortunately our interview was cut short, due to a fire alarm at the conference center. We had to exit the venue, the irony being that when we got outside, there were lots of protesters picketing the venue about the carbon emissions from Data Centers. These same protesters were recording the mass exit, obviously to be uploaded onto any and all platforms and shared with many, many people. But where is that data stored? Yes, you guessed it, that data is stored within the Data Centers causing all the controversy for their power usage. It's clear we need to find a better solution to help Data Centers achieve their aim, to be climate neutral, and we think open source hardware and software can help get them some of the way there.
Glossary of Terms:
- OCP: Open Compute Project
- ONF: Open Networking Foundation
- TIP: Telecom Infra Project
- Bare metal: A piece of hardware without software pre-installed
- Stratum: Stratum is an open source silicon-independent switch operating system for software defined networks.
- CORD: Central Office Re-architected as a Datacenter. CORD is a platform which leverages SDN, NFV and Cloud technologies to build agile datacenters for the network edge.
- VM: Virtual Machine
- Alt-Net: Smaller independent ISPs focusing on specific, localized regions
- OLT: Optical Line Termination
- vBNG: virtual Broadband Network Gateway
- Open RAN: Open Radio Access Network