Published Dec 6, 2019

Amir Michael

Amir Michael delves into the transformative power of the Open Compute Project on infrastructure technology, shares his insights from Google's groundbreaking server and data center designs using consumer-grade components, and explores the challenges of firmware in data centers, emphasizing innovation through collaboration and cost-efficiency.
Episode Highlights
On The Metal logo

Popular Clips

Episode Highlights

  • DIY Servers

    Google's early server design was marked by a DIY approach, utilizing consumer-grade components to avoid the high costs of enterprise hardware. describes how they sourced parts from local retailers and assembled them on makeshift racks, emphasizing cost-effectiveness and ease of maintenance 1. This approach included innovative solutions like using Velcro to secure components, which facilitated quick repairs and replacements 2.

    The joke was they would buy dram that was swept off the floor at the factory because it was cheaper and it had more errors.

    ---

    This philosophy of using inexpensive, unreliable hardware required robust software to handle errors, marking the beginning of Google's scale-out hardware strategy 3.

       

    Rack Design

    The evolution of rack design at Google was driven by the need for efficiency and scalability. explains that the initial designs aimed to maximize server density, reducing costs by spreading infrastructure expenses across more servers 4. This led to innovative rack configurations, such as the three-column design, which optimized space and power distribution.

    We said, let's do what makes sense and not try and fit it into any constraint that was defined by the facility.

    ---

    However, challenges in packaging and shipping these racks prompted further refinements, resulting in the development of the Open Rack standard, which balanced cost and practicality 5.

       

    Container Data Centers

    Container-based data centers offered a novel approach to infrastructure deployment, though not without challenges. recounts the initial enthusiasm for using shipping containers to house servers, which promised flexibility and rapid deployment 6. However, logistical issues and increased costs led to the abandonment of this model in favor of more traditional data center designs.

    The container made that loop very short, and so you spent a lot less energy on moving air.

    ---

    Despite this, the principles learned from the container experiments, such as efficient cooling and power distribution, were integrated into future designs, enhancing overall data center efficiency 7.

       

    OCP Revolution

    The Open Compute Project (OCP) revolutionized data center infrastructure by promoting open-source hardware designs. highlights the project's smooth internal adoption and its impact on the industry, encouraging collaboration among major tech companies 8. This initiative allowed companies like Facebook to develop cost-efficient infrastructure tailored to their needs, reducing reliance on traditional vendors.

    We need something more cost efficient. Come in and figure it out.

    ---

    The OCP's collaborative approach fostered innovation, leading to more sustainable and scalable data center solutions 9.

Related Episodes