XFaaS: Hyperscale and Low Cost Serverless Functions at Meta
In this paper from the 2023 Symposium on Operating Systems Principles, the authors describe Meta's internal system for serverless called XFaaS, which runs trillions of function calls per day on over 100,000 servers. The paper discusses the challenges the authors faced in building and optimizing the infrastructure, such as handling load spikes, ensuring fast function startup and execution, global load balancing, high utilization of resources, and preventing overload of downstream services. The architecture of XFaaS consists of five main components: Submitter, load balancers, DurableQ, Scheduler, and Worker Pool. Clients schedule function execution with the Submitter, which interfaces with downstream components. The function execution request is forwarded to a load balancer, which then stores the function information in DurableQ. A scheduler determines the order of function calls based on criticality, execution deadline, and capacity quota. The scheduler assigns function executions to the Worker Pool using a load-balancer approach. XFaaS also includes performance optimizations such as time-shifted computing and cooperative JIT compilation. The paper evaluates XFaaS's ability to maintain high utilization, efficiently execute functions, and prevent overload of downstream services. It concludes by discussing the uniqueness of XFaaS in characterizing a serverless system at immense scale and raises questions about whether certain functions would be better served as batch jobs and the reliability of anecdotal knowledge on XFaaS utilization.
Intel CEO Pat Gelsinger is facing challenges as the company’s struggles continue, with its stock falling more than 12% following a disappointing Q1 projection for both sales and profit. Demand has weakened in the data centre processor market, and other businesses such as programmable chips and components for self-driving vehicles are also struggling. Additionally, Intel’s attempts to enter the foundry business have not yet taken off. Gelsinger acknowledged that the first quarter has not gone as well as hoped, but he expects 2024 to improve. Intel has been making efforts to compete with chips manufactured by Advanced Micro Devices (AMD) on TSMC’s process and also faces competition from ARM and GPUs.
Why Custom GPTs are better than plugins
OpenAI has announced the introduction of Custom GPTs, with over 3 million GPTs created since then, and over 200,000 publicly available. Unlike traditional plugins, Custom GPTs allow for pre-configured actions and capabilities, as well as embedded knowledge specific to the task at hand. This enables users to switch conversations depending on the task, making it quicker and more intuitive. The benefits of Custom GPTs over plugins are numerous. Firstly, GPTs are open to everyone, with the ability to be listed on the "GPT Store" or shared via direct URL. Secondly, adding knowledge to a GPT is as simple as uploading a file, whether it be a PDF, raw text, or markdown. Custom GPTs also offer various capabilities such as web browsing, DALL·E image generation, and code interpretation. Furthermore, GPT creators can provide conversation starters to give users suggestions on how to use the GPT effectively. Unlike plugins, GPTs do not require coding or servers, as the necessary context and knowledge can be added through instructions and file uploads. Lastly, OpenAI now supports mentioning GPTs within conversations, allowing for enhanced functionality. Overall, Custom GPTs offer a more accessible and versatile approach to development.
Square Kilometre Array prototype 'scope achieves first light
The SKAMPI radio telescope, designed by a team from ten different countries and built in China, has achieved first light. The telescope is one of the designs for the Square Kilometre Array (SKA), a massive international project to build the world's largest radio telescope. The SKAMPI model was assembled in South Africa in 2018, where it will be a part of the SKA. Testing began in 2019 and the technical commissioning work was completed in early 2022. The SKA team has since worked on enabling robotic operations and developing Docker-based software tools for the telescope. First-light observations with SKAMPI have already been successful, with imaging of the Southern Sky and detection of the Vela pulsar. The SKA organization is pleased with the results, as it shows that SKAMPI is working as expected. The completion of SKAMPI will inform the development of the larger SKA-Mid telescope, while the SKA-Low facility will be built in Australia. These projects will require significant compute and storage infrastructure to handle the large amount of data they will produce.
Show HN: Open-source x64 and Arm GitHub runners
Ubicloud is a unique cloud platform that offers an open, free, and portable alternative to traditional cloud providers. Similar to how Linux provides an open alternative to proprietary operating systems, Ubicloud aims to provide users with greater control and transparency over their cloud infrastructure.
One notable feature of Ubicloud is its open-source nature, allowing users to access and review the source code on GitHub. This fosters a collaborative environment and enables users to contribute to the platform's development. Additionally, Ubicloud offers managed runners for GitHub Actions, making it convenient for developers to use within their existing workflows.
Another advantage of Ubicloud is its use of cost-effective bare metal instances for its managed service. By utilizing these bare metal instances, Ubicloud can optimize costs and pass those savings onto users, making it an attractive option for those seeking a more affordable cloud solution.
In summary, Ubicloud provides an open, free, and portable cloud platform that promotes transparency, flexibility, and affordability for users.