Distributed SQL Summit Recap: The Future of Data Infrastructure, A Telco Roundtable

January 14, 2021

At the Distributed SQL Summit 2020, leaders and architects from notable enterprises in the telco industry joined us for a panel to discuss the future of software development, including important trends such as IoT, 5G, Edge, cloud, and their impact on data infrastructure. Our panelists included Puneet Devadiga from Rakuten Mobile, Hale Donertasli from Rakuten Mobile, Kartik Rallapalli from TracFone Wireless, and James Taylor from Comcast; the panel was moderated by Yugabyte’s head of Data Engineering, Prasad Radhakrishnan.

In this blog post, you’ll find a summary of the panel discussion. If you want to watch the panel session, here’s the video playback: The Future of Data Infrastructure, A Telco Roundtable.

Setting the stage for the panel discussion, Prasad noted that telcos, now more than ever, play a very important role across the world. Mobile and internet are foundational to nearly everything we do in a COVID world.

With the foundation laid for the critical role of telecommunications today, Prasad poses this question to the panelists:

What are some of the top innovations (in telco) that excite you the most?

Hale kicks off the discussion by articulating these exciting innovations: CUPS for control plane and user plan separation, as well as COTS to build common cloud infrastructure without the need to buy vendor specific hardware. The most important innovation in her opinion was organizations taking a cloud native approach to network functions.

Next, James explains that at Comcast, a lot of their innovation is focused on continuing to expand bandwidth to stay ahead of the load coming from an increasing number of smart, connected devices. In addition, personalizing the customer experience and making that personalization global, is an area of innovation. James notes: “If you happen to be on a trip in the Bahamas somewhere, you still want to know that how you access your products and services, whether it’s checking on home security and so forth, that those tools and those products are still working and you’re able to interact with them no matter where you are in the world.”

Kartik then made the point that there will be also be innovations in IoT, virtual reality, and artificial intelligence, so staying agile as an organization is essential to staying ahead of those technology trends. In addition, it’s also essential to be able to scale and be resilient.

Finally, Puneet dove into 5G, explaining that low latency, resiliency, and portability are essential for 5G. Getting closer to the end user means using edge data centers, which don’t have massive computing power; in contrast the servers can be quite small. Being able to run Kubernetes deployments with telemetry and management workloads on small machines at the edge is an innovation, as well as the automation that goes along with that.

Continuing with the discussion, Prasad asked the group:

What kind of impact is COVID having on your business?

James began by explaining that internet traffic patterns have changed at home. Before COVID, if people went to an office location or school during the day, there was no real internet traffic at home. But now, home internet needs to be always on and always high.

Kartik agreed with James’ points about increased usage from people across the globe. He also notes that online sales are increasing because of social distancing practices, and digitization efforts are also increasing as more activities are brought online.

Next, Puneet explained that a year ago, Rakuten Mobile introduced the concept of zero touch provisioning, so it was possible to bring up an entire site, virtually, within 10 minutes once connectivity was in place. The idea that engineers don’t need to go on site for build up activities was already in place, which has helped in a COVID world. Additionally, working from home has been a cultural shift. In person whiteboarding sessions and interactions are gone, and all meetings are now virtual; but the team is coping and well prepped for success in a virtual world.

James added that the investments made over the years to move a cloud native architecture with modern technologies is now being seriously tested, and it has been exciting to see the modern environment hold up.

Finally, Kartik noted that at TracFone, automation is also important; complete automation across DevSecOps, infrastructure, transactional systems, data, business, operations, finance – across the board.

Prasad let the panel know that the discussion would later segue into apps and data, but will start with a discussion around cloud.

Cloud seems to be a top trend for most organizations, and, would you agree that broadly your time and resources are split between two categories – legacy and modernization efforts vs. new next generation applications? Which one do you think is more “needy,” in the sense that you spend more time on one vs. the other? What’s the impact of those initiatives on your network and the database layer because of it?

Hale kicks off by stating that cloud is not a trend, it’s a must. From a telco perspective, cloud native is one of the key enablers of 5G. A lot of innovation is happening within all the layers of the network to make 5G possible. Staying true to the CAP theorem, because massive partitioning is happening in layers of the network, the databases need to be unified.

Puneet then builds on this, stating that his team is building a Kubernetes based platform for Rakuten Mobile, and that this platform layer is being offered to all applications, even legacy applications when they are modernized and ready.

Kartik explained that TracFone is modernizing, too, moving away from monolithic implementations to more horizontally scalable implementations. Innovation is happening in internal tooling, too, around automating release management, CI/CD, and DevSecOps capabilities, and there are plans to release some of these tools as open source.

James explained that “legacy” is a loaded term. Some of the time the legacy app “just works”. But other times, you run the application code through SonarQube and find that there are 365 days of technical debt, giving you an indication of the quality of the app and how long it would take to refactor. So, at Comcast, they took a strategic approach to migrating to cloud native architecture. They began slicing the monolith based on business strategies, like decoupling services powering high revenue generating products vs lower generating ones. (“The first cut is like cutting a diamond; the first cut is your most important cut when it comes to decoupling the code.”) Once the monolithic application was broken into microservices, they began looking at data services and are now looking at how to get the data distributed in a way that’s more resilient and flexible, too.

In conclusion, Kartik explained that at TracFone, modernization involves a change in organizational mindset, too. “Design to survive failure at each and every point has to be embedded into the implementation architecture and design. That’s one of the challenges we are addressing, and we are handling that through design workshops as well as measuring the design maturity from an agile point of view.”

Continuing the discussion, Prasad asked the group:

What is cloud to you? And is Kubernetes an important part of your cloud strategy, and why?

Puneet answered first by summarizing that most definitions revolve around infrastructure and the virtualization layer, and then he builds on the idea, stating that what adds value are the services you build on top of it. By services, he clarifies that he’s not talking about the applications that are onboarded, but rather the services that make the life of a developer easy. At Rakuten Mobile, the application platform for company developers offers services out of the box, like resiliency, fault tolerance, scalability, and elasticity.

James continued the cloud theme, noting that for startups cloud is a way to accelerate building the application. But for a large organization that has invested tens of millions of dollars in data centers, cloud helps support the overall application architecture, with services such as messaging or an API gateway. James acknowledged the trend to multi cloud, where organizations need multiple cloud providers, to choose between providers with a location nearest users globally, and technologies that run across clouds. James concluded by saying Cloud Foundry has been instrumental in driving developer experience, along with Spring Boot and the Spring framework, and that Kubernetes comes into play in helping drive and support other use cases where you need data, storage, and persistence, and a network control plane with Istio or Envoy.

Kartik added that cloud enables the highest level of agility to deploy infrastructure with resilience and scale. And with Kubernetes, organizations like TracFone can more easily migrate workloads from data centers to AWS, for example. Kartik warned though that public cloud is becoming more segmented with regards to implementation practices, so Kubernetes takes on a role in helping standardize the developer experience.

Continuing on the Kubernetes theme, Prasad followed up with this question to the panel:

Where do you see this going, and what kind of problems do you think this can solve in a better way than what we had before in terms of running these stateful database workloads on Kubernetes?

James answered that Kubernetes is extremely distributed, can monitor itself and restart crashed containers when needed, and is an operating infrastructure platform that enables you to migrate workloads across public cloud environments. He concludes by saying running databases on Kubernetes is a “really good use case for the platform.”

Puneet continued by stating that Kubernetes is adding value in terms of being able to streamline day to day operations, provide the ability to do graceful upgrades, and brings in automation which helps “manage the lifecycle with very little or no impact to the end user.” It’s attributes like these that Kubernetes is bringing to databases and applications.

With the conversation naturally heading into the next question, Prasad proceeded to ask:

How big do you think the scaling challenges (are)? Any thoughts, and how do you plan to address it, more so on the stateful layer?

Hale explained that high scale is really difficult to test; while there are methodologies, testing to see if a system really can scale to desired levels does present a challenge. And from a stateless perspective, Hale shares this formula: stateless + stateful = stateful. If 70% of your apps are stateless and 30% are stateful, the whole system can be considered stateful. So, Hale advises to make everything stateless.

Kartik added that at TracFone, they want microservices to be completely stateless, and at the same time, they need a data layer that is highly resilient and can scale.

James brought in the concept of separation of the data plane and the database service itself. If you’re looking at a distributed database, those nodes can go up and down, and because they are handling requests for data, and the data is sitting in a managed network provisioned storage environment, the requests can be handled with extreme low latency.

Prasad summarized that the panelists have their application architectures finely tuned, and seem to be moving into solving data challenges as a next frontier. Prasad then asks the group the final question:

How do you see Distributed SQL as a notion, as a concept, the architecture, and how does that come together in trying to solve your challenges of today and tomorrow?

Hale started by saying “it is actually mandatory for 5G” because of the ultra low latency of one millisecond for some services and “therefore we need to move the core network functions closer to the user now.” Distributed SQL comes into the picture because instead of all queries going to a central location with attached delay, some queries need to be handled at the edge, which distributed SQL can do.

Puneet then added that database needs have evolved; they started with a simple active/standby setup, then went to active/active, and the types of conversations today involve geo replication.

James then took us back to 2014 when Comcast started their cloud native journey, and there weren’t really any cloud native databases out there, unlike today. They started going to VMs, then containers, and the applications were modernizing, but not the databases. Now, they are evaluating databases and saying “we don’t want active/active, we don’t want active/passive, we want something that basically is literally just always on.” James gives YugabyteDB a shout out for being a database that’s built from the ground up to be cloud native and uses parts of Postgres, one of the most solid database platforms out there.

Kartik agreed that there has been a lot of innovation in the application development side, but very little on the database side until now. Distributed databases are evolving, and it’s helping with microservices architectures. Kartik gives Postgres a shout out for being a great database because it supports both SQL and NoSQL through the JSONB datatype. He adds that “in our organization, all our developers know SQL, all our business users know SQL…so leveraging an existing framework for developing a distributed database architecture is something very beneficial for the organization to adopt.” Migration will also be a lot easier from a monolith to distributed SQL because you can migrate the schemas and follow ANSI SQL standards.

Prasad wraps up noting that the panelists are all driving mission critical systems and new initiatives across their organization, and thanks James, Hale, Puneet, and Kartik for their time and for sharing their views.

🙏

To get the complete panel experience, check out the recap video: The Future of Data Infrastructure, A Telco Roundtable.