Lightbend

Kifi CTO and co-founder Eishay Smith faced a significant challenge in building an application capable of performing multiple computationally intensive tasks within a limited time frame. The architecture needed to support high performance at all scales and be flexible enough to scale up and down instantly to meet unpredictable customer usage patterns. Kifi required blazing fast response times throughout the application tiers, robust support for multithreading to maintain performance, and support for rapid continuous deployments. Additionally, the application needed to achieve near-zero downtime and 100% resilience in case of server faults. A strong Service Oriented Architecture (SOA) orientation in the underlying software languages and frameworks was also critical.

The Royal Netherlands Marechaussee faced the challenge of enhancing border security following the Schengen Agreement, which eliminated many internal border controls in Europe. To address illegal immigration and cross-border crime, the KMAR needed a system that could efficiently collect and analyze anonymous data, observe vehicles, and respond to quick alerts in emergency situations. The system had to be capable of processing data from various sensors and cameras in real-time, ensuring that potential suspects could be identified and stopped promptly. The complexity of aggregating real-time data from multiple sources into a single view of a vehicle, while maintaining high precision and speed, was a significant challenge. The system also needed to be highly distributed, asynchronous, and capable of handling numerous border checkpoints and patrol cars.

In the OEM sector, demands for speed, quality, and cost control are rising. OEMs are targeting higher levels of automation for their production lines to meet stringent client requirements around just-in-time and just-in-sequence product deliveries without increasing operational costs. However, automating an entire production line with multiple systems from different vendors presents challenges. Unomic has been developing the ELFiNOS platform to support larger and more diverse clients in the manufacturing sector as edge computing and IIoT technologies mature.

SoftwareMill and Yap.TV began working together in September 2011. At the time, the Yap.TV server was entirely Ruby-based. While this worked in the beginning, as the user base grew, high load on the servers, especially during prime time TV, caused significant issues with server performance and scalability. Not only was this degrading the end-user experience, but as new features were waiting in the backlog to be developed, key engineering resource time was becoming dominated by scaling servers and operations. Apart from serving traditional web traffic, Yap.TV also required an increasing number of background imports, taking together data from various sources. To handle the ever-growing amount of data to be processed, a performant (but at the same time - easy to program and safe) concurrent programming model was crucial.

Walmart Canada aimed to enhance its online store experience by creating a responsive web presence across various devices while reducing backend costs and improving scalability. The existing architecture based on Oracle ATG required specialized development teams and long cycles, which hindered development speed. Additionally, the reliance on expensive hardware limited scalability and performance, as the legacy Java infrastructure could only handle limited simultaneous requests. Walmart needed a modern, event-driven architecture to handle more concurrent processes without additional compute capacity. Integration with existing infrastructure was crucial to avoid downtime, as the ecommerce store needed to remain operational at all times.

Traditional paper notebooks are becoming obsolete, and digital note-taking is on the rise. However, a significant challenge with digital notes is that users often forget to revisit them, leading to important information being overlooked. Digital notes, while convenient, can become lost in a sea of text, making it difficult to extract actionable items, phone numbers, and calendar entries. This problem is exacerbated by the fact that digital notes are often stored in the cloud, making them 'out of sight, out of mind.' Tindr's CEO, Mike Kelland, recognized this issue and sought to develop a solution that would address the shortcomings of digital note-taking.

Ensuring that subscribers have access to only the audio/video content they have purchased or subscribed to is an ongoing challenge faced by cable companies worldwide. The problem is further compounded as cable companies offer various promotional bundling packages from time-to-time that changes consumer entitlements on a regular basis. CCAD has solved this problem in an interesting way with a product suite of several different products, the largest one named CASMR. This product set - a combination of both hardware and software - is utilized by cable companies worldwide to solve the problems with managing entitlements, conditional access and integration with billing systems. It’s important to point out that these are predominantly “lights-out” systems that actually run the cable company and are used by employees who are trying to effect some change in the system. Subscribers today have more choices therefore forcing providers to constantly adjust content offerings and prices. In addition, technology and consolidation trends have enabled service providers to significantly grow their subscriber base. The subscriber experience must be flawless from both a timing and accuracy perspective regardless of subscriber population.

Tubi, an ad-supported video on demand service, was facing challenges in scaling its systems to meet the growing demand and improving user experience. The company needed a platform that could provide a responsive experience, automatically scale on demand, handle the complexity of a video on demand service, and integrate with existing technologies like gRPC, ScyllaDB, Postgres, Datadog, Spark, and AWS. Additionally, Tubi wanted to leverage big data and machine learning for hyper-personalized content recommendations and enable developers to focus on business logic rather than maintenance tasks. The solution needed to be implemented quickly to keep pace with Tubi's fast-paced innovation and business goals.

After 7 years of PHP development, Gawker Media faced challenges in maintaining a large 300,000 line PHP codebase. Additionally, managing two separate production environments (LAMP and JVM) was making development inflexible and inefficient. Gawker maintained a development team split between New York and Budapest, with each team responsible for one technology. The current systems were built on a mix of Java and PHP, which made it difficult to share work across teams due to a lack of expertise. Moving to a common platform was necessary to concentrate development efforts on building the application instead of dealing with multiple environments and deployment processes. The new platform needed to provide a solid base to accommodate aggressive growth requirements. Gawker set upon evaluating the technology marketplace, aiming to leverage their experience with JVM-based solutions and ensure interoperability with existing Java components and libraries.

WhitePages faced significant challenges with its existing service-oriented architecture, which relied heavily on legacy languages like Perl and Ruby. These languages were not only costly to scale but also inefficient in terms of hardware utilization. The company was running over 80 servers to handle 400 QPS, leading to high hardware costs and significant latencies due to single-threaded applications. The existing setup also required extensive serialization and deserialization of data, consuming 70% of server time. WhitePages needed a more efficient, scalable, and reactive development environment to reduce hardware costs, improve throughput, and enable rapid development iterations.

The goal of Primetalk is to create a completely new stack that allows the creation of flexible user-initiative dialogs. However, this goal is challenging due to several factors. Natural Language Processing (NLP) is a complex field requiring interdisciplinary knowledge in artificial intelligence, neurophysiology, aural physiology, signal processing, computational linguistics, probability theory, and mathematical statistics. Additionally, automatic, large vocabulary continuous speech recognition is a heavy load task that benefits from parallelism, but implementing parallelism is non-trivial. The dynamic event-driven nature of real-time dialog handling presents another challenge, as many events need immediate handling by the dialog logic. These events can include recognition results, audible user interventions, timers, telephony, and system playback. Furthermore, simultaneous playback of different audible media and parallel conversations with multiple participants add to the complexity. Scalability is also a concern, as real-world demand requires handling many channels by a single application instance with zero administration. The initial version of the Primetalk Speech Portal, developed with Java concurrency, faced several drawbacks, including memory leaks, error-prone communication between threads, limited scalability, and unexpected runtime errors due to incomplete configurations.

Klout faced a significant challenge in expanding its services into the mobile application market. The existing infrastructure, which was primarily based on PHP servers and Java-based applications, needed modernization to support the new mobile applications and partner-based applications. The key goal was to unify the platform and upgrade the technology to ensure a guaranteed level of performance as the platform scaled. This required re-architecting the infrastructure to meet scalability and consistency demands. The engineering team had to find a solution quickly due to aggressive delivery schedules, and they needed a platform that could handle real-time data aggregation across several social networks and data stores.

Prior to 2012, Sharethrough's advertising platform relied on an integration between their end-to-end campaign trafficking and analytics platform and a third-party ad server. To realize their vision of powering native advertising across the open web, they needed to build their own ad server as no existing solution supported the targeting, templating, and analytics capabilities they required. Sharethrough started small, focusing on building early pieces of the core templating and targeting technology as a simple, horizontally scalable trafficking tool in Sinatra, deployed on Heroku. As the platform quickly gained traction, they realized that a move from Sinatra (and Heroku) was going to happen sooner than originally planned. Given the significant performance requirements and the critical portion of their infrastructure being in the service layer, Sharethrough decided to examine other languages and frameworks.

LinkedIn faced challenges with traditional Java web frameworks, which were time-consuming and inefficient for developers. The existing frameworks required a lengthy process to make changes, reload, and test, leading to wasted developer time. Additionally, stateful, thread-per-request frameworks performed poorly and were difficult to scale, resulting in inefficient use of cloud resources. While LinkedIn used some non-Java platforms like Ruby on Rails or node.js, the majority of their server-side code was written in Java. Therefore, they needed a modern UI framework that could cater to Java developers and improve their productivity.

HolidayCheck faced significant challenges in scaling their platform to support new international sites. The existing LAMP stack was becoming increasingly complex and unstable, particularly when trying to duplicate code for different markets. The platform's performance and stability were major concerns, and the need for a new architecture was evident. Christopher Schmidt, the Senior Lead Architect, was tasked with designing a new technology stack that could support high-performance websites and dynamic cloud deployments. The goal was to deploy new or changed front-end applications within days, a challenging objective given the complexities involved.

Flowdock faced the challenge of creating a low-friction collaboration tool that could act as a shared team inbox with group chat capabilities. The goal was to enable teams to stay up-to-date, react quickly, and integrate various project management tools, version control systems, and customer feedback channels into a single, easily consumable stream. The core-messaging module needed to be highly performant and scalable to handle substantial server-side infrastructure traffic. The team initially considered using Ruby on Rails, but the core-messaging module required a more robust solution. Node.js was considered but had limitations, and Erlang lacked Java interoperability. The team was interested in the Actor model, which led them to explore Scala and Akka from Typesafe.

Heluna faced a significant challenge in scaling its anti-spam service to meet the growing demand for processing millions of email messages daily. The existing platform, which relied on legacy technologies like Sendmail, Perl, and C, was unable to scale horizontally to accommodate the increasing volume of emails. This lack of scalability threatened the reliability and performance of Heluna's service, prompting the need for a more robust and scalable solution. The founder and chief architect, Mark Beeson, recognized that the current architecture was untenable for long-term growth and sought to improve the system's scalability, performance, and operational visibility.

Abiquo faced a significant challenge with its cloud provisioning platform, which was unable to handle unpredictable scaling demands effectively. The system's massive configurability, initially a strength, became a bottleneck as it hit peak throughput at relatively low loads. The central issue was the synchronous calls made by the Abiquo Node to the Virtualization Factory, which handled the configuration of end-users' virtual environments. These calls were slow, creating a bottleneck and leaving users without status updates during the creation of virtual environments, leading to confusion and dissatisfaction.

Valtech’s client, a UK broadcaster, faced the challenge of delivering daily insights to approximately 2000 stakeholders across the business. The broadcaster aimed to transform their monthly in-depth reports into a more frequent, easily consumable format that could be accessed on any device. The challenge was compounded by tight deadlines, aggressive targets, and a fixed budget, necessitating a solution that was both efficient and scalable. The broadcaster required a platform that could handle extensive data manipulation, including sorting, grouping, mapping, and combining data from a NoSQL data store and existing REST APIs. The solution also needed to generate bespoke SVG visualizations on the server, which would be presented through a responsively designed website and email channels. The broadcaster turned to Valtech to bring this vision to life, seeking a technology platform that could meet these complex requirements while ensuring ease of use and maintainability.

Conspire faced the challenge of revamping their backend system to support a customer-facing product. Their original backend was a multi-threaded Java application with a traditional concurrency model, which was too complicated and not suitable for future developments. The team needed a new solution that could handle scalability, resiliency, and simplicity, as the existing codebase was not up to the mark.

John Deere wanted to deliver increased return on investment to customers purchasing their large industrial equipment. The challenge was to make sense of the immense amount of data generated by the sensors embedded in their industrial combines and tractors. These sensors capture data on various operational aspects, including hydration, fertilizer, and pesticide levels, as well as the efficiency of machine operation. However, without proper analytics, this data holds no value for farmers in their decision-making process.

Gilt Groupe faced challenges in scaling their e-commerce platform, originally built on Ruby on Rails, to handle high traffic volumes and provide unique shopping experiences. The need for a more scalable and efficient solution led them to explore alternatives to Java, eventually settling on Scala due to its productivity and ease of integration with legacy systems. The introduction of Gilt Live, a real-time shopping experience, required specialized technology like WebSockets, prompting a reevaluation of available web frameworks.

To bring its offering to life, StashAway targeted a way to aggregate thousands of individual customer trades into one, and then to immediately distribute the share units and funds back to each customer as soon as the trade is executed and settled. StashAway’s capital markets license requires the company to carry out these settlement processes on time, every time, as defined by the rules of the Monetary Authority of Singapore and the respective regulatory entities in the other markets in which the company operates. The penalties for non-adherence to these rules are severe: from fines to the suspension of operations or even the complete revocation of the company’s license to conduct business. As StashAway began to develop ERAA, it recognized that its choice of technology platform had the potential to make or break its bold vision for industry disruption.

Managing network devices is cumbersome and outdated, requiring IT professionals to log into each device via a command line interface (CLI) to make changes. This approach is analogous to outdated mainframe practices and is unnecessarily low-level and painful. The complexity of managing a network infrastructure without a dedicated team of networking engineers can lead to inflexible networks that struggle to keep up with the latest trends. IT professionals are expected to manage high-level requirements down to machine code, which is a challenging proposition. The fear of unknown repercussions makes IT professionals hesitant to make changes to the network.

The Guardian faced the challenge of transitioning from a print-based organization to a digital-first entity. This required modernizing their existing web infrastructure, which was based on a heavily customized Content Management System (CMS) that was difficult to maintain and extend. The existing platform, although functional, was a monolithic application that slowed down the process of adding new features. The Guardian needed a more agile and scalable solution to support their digital-first strategy and to keep up with the fast-paced demands of modern web development.

When Tomax Corporation launched Retail.net over a decade ago, it was the first technology and solutions provider to take the SaaS route and pursue a true demand-driven retail system that could scale up and scale down to meet the needs of shoppers, store associates, and suppliers alike. As the mobile revolution took the retail world by storm, Tomax realized it needed to revamp its core technology platform to deliver a far more flexible and extensible suite of offerings to keep customers happy. Retailers were rapidly moving away from purpose-built POS and inventory management devices towards software built on standard smartphones and tablets. Shoppers, too, were increasingly accessing retailer’s systems, either in-store or outside, from mobile devices. Similarly, the rise of multi-channel retail, both as a sales channel and as a management and information vector, necessitated that Tomax rethink its core technology offering. At the same time, the mainstream adoption of cloud computing called into question the old enterprise practice of overprovisioning IT infrastructure to meet peak demand, which for retailers can be as much as 50x depending on seasonality. Lastly, Tomax felt that it needed to take the next step up from Java and towards one of the newer programming languages better equipped to handle the massive concurrencies of mobile environments and the scaling flexibility required to properly leverage the new world of public, private, and hybrid cloud computing.

Dream11, India's largest fantasy sports platform, faced significant challenges due to its rapid growth from 1 million users in 2014 to over 80 million by 2019. The platform's original application architecture struggled to handle the massive spikes in user traffic, especially during peak times leading up to real-world sporting events. This resulted in users being unable to register for contests or being kicked out of the registration process. Additionally, the system strained to ensure contests did not exceed the maximum number of participants, leading to cancellations and refunds, which weakened customer loyalty. Dream11 required a highly elastic, self-healing system that could scale with increasing demand, integrate seamlessly with AWS, and reliably adapt to predicted growth. The challenge was to implement this solution before the Indian Premier League season, a peak period for Dream11, to avoid significant revenue loss.

HPE aimed to enhance customer value by delivering insights in seconds and minutes rather than hours or days. InfoSight, a leader in infrastructure monitoring, needed to evolve from its batch mode, big data architecture to a real-time analytics platform. The challenge was to process the massive amount of data from over 20 billion sensors globally, which send trillions of metrics daily. The existing batch processing was too slow, and a new approach was needed to deliver insights faster.

Dealing with healthy growth in product usage is a problem that every company wants, but it can be challenging to provide an efficient and consistent experience for every user. The Lucid engineering team could see that their monolithic server architecture had too much overhead and some limitations that would make it difficult to scale in a cost-effective way. They realized that they needed to make a marked shift in technologies to ensure that the company could meet its performance, scalability, and reliability goals. Lucidchart’s growing pains were due to both architectural realities and to the limitations of certain technologies. For example, Lucidchart’s monolithic application was not easily partitioned and distributed, and CakePHP introduced a lot of overhead per request, resulting in the minimum response times being too high under load. As it became clear that a significant portion of the code base needed to be rewritten, Brian Pugh, VP of Engineering, pushed for a complete survey of existing solutions to find the best tools to meet both current and future technical requirements. The team determined that scaling up in a cost-effective way would be best achieved by adopting a Service Oriented Architecture (SOA) and making significant use of parallel processing to handle compute-intensive operations. Building such an architecture from the ground up is a significant task, so they wanted to leverage existing technologies and frameworks that were purposefully built for the task. The Lucid team narrowed in on PHP, Java, and the Typesafe Platform as possible technologies on which to base their future development.

The existing platform was a huge monolith, and since the code base was so large, software builds often took a whole night to run. Setting up a testing, development or production environment required five to 10 days of work, and deploying an emergency fix to production took 24 hours. Even worse, the legacy platform couldn’t handle the massive increase in site traffic on holiday sales days and other big events, such as the launch of a new iPhone. So, for more than a decade, Verizon’s engineers had to recreate the entire e-commerce site in parallel with a special light code base that disabled logging, removed all complex transaction requirements, and was subjected to massive load testing. This bespoke website only ran in production for a few days each year, even though it took more than six months to create. Verizon’s site never crashed on the holidays or after a new iPhone launch, an industry record unmatched by any other cell provider. But all of that special code was tossed out every year and never re-used.