Exclusive: DCW Keynote Interview With Rebecca Weekly of Cloudflare

Ahead of her keynote panel at Data Center World 2023, Cloudflare's Rebecca Weekly talks with Bill Kleyman about talent shortages, diversity, and sustainability in the data center industry.

Bill Kleyman

April 17, 2023

6 Min Read
Photo of attendee at the Data Center World 2022 conference and exhibition.
Data Center World

Ahead of her keynote panel at AFCOM Data Center World 2023 in Austin, we met with Rebecca Weekly from Cloudflare to get her perspective on our connected industry. Weekly leads the organization that builds Cloudflare's compute, network, and storage hardware systems. Together, they shape Cloudflare's development strategy to meet internal and external customer needs as they deliver over 20% of the world's Internet traffic.

At Data Center World this year, Weekly's fireside keynote will explore how data, connectivity, and new compute-heavy workloads (think ChatGPT) are changing and modernizing data center infrastructure. In her keynote, we'll explore how leaders can prepare for a more connected, data-driven future.

Be sure to join us for a fantastic session!

Data Center Knowledge: How would you describe the current state of the talent gap that has impacted the data center industry? What steps can the technology industry take now to mitigate this labor crunch?

Rebecca Weekly: With the current economic situation, we have been seeing an easing of the "talent gaps" since around December of 2022. Fundamentally though, this is a long game -- we must continue to educate and develop talent to ensure that our talent pools don't fluctuate with the economic conditions, up to and including leveraging increased automation and AI to automate any tasks possible.

Related:Data Center World 2023 Keynotes: What to Expect

Fundamentally, too, much infrastructure education (predominantly DevOps) has been assumed by large cloud providers (AWS Certified, Azure Certified, etc.) versus neutral strong sources looking across the domain space with an eye toward how to optimize for workloads themselves.

Increasing communities are trying to step up here (Linux Foundation, Open Compute Project, MOC Cloud Alliance, etc.) and universities, of course, but this is a significant gap for creating a diverse and thoughtful talent pool. We need significantly more debate and discussion versus indoctrination to various vendor-centric approaches if we want a growing and discerning community of developers. Then again, most businesses care about time-to-results versus learning, so I believe this has to come from developers and academic institutions themselves versus assuming that any vendor will truly sponsor this work.

Weekly: AI is the easy on here: increased automation of knobs/tuning for optimal performance and power consumption based on dynamic load conditions, automation of DC "healing" (reboots, firmware updates, etc. to resolve server and network issues without requiring direct human interaction), dynamic workload placement, generation of testing suites based on high-level descriptions. None of this is new per se, but it is clearly exploding. Even more interesting is how we see silicon development starting to leverage ChatGPT.

Related:Data Center World 2022: Postcards From Austin update from May 2022

Of course, this, too, was foreseen by AI gurus such as Jeff Dean (although that was a specific time-consuming issue associated with manufacturable layouts versus generative AI for the higher-level logic). Other domains where I see key trends specifically relevant to operators are around sustainability: active power, temperature/performance throttling for better dynamic use of power, green power sources, better heat reuse methodologies. We as an industry must invest here, and it is not just a Silicon and Systems investment, it requires standardization of management and facilities so we can create better commercial incentives to align with sustainable use cases.

For example, many contracts are flat for space, power, and connectivity regardless of the actual usage -- this creates no incentive to optimize utilization of power by load and means no company will invest even though that would be ideal for the environment. We need to change these economics and reward customers for their optimizations to see real change. This may be a place where regulation (like we saw on home devices with power idling requirements when appliances were not in use) could be valuable.

DCK: What should the data center industry do to encourage and bolster a more diverse workforce?

Weekly: There has been much written about this -- diversity comes from a host of sources (background, education, gender, race, religion, culture), and the most definitive research I have seen is that the only way to change the outcome is to break out from "who you know." Most companies reward recommendations, which indicates that people who have worked together before will continue to work together and will inevitably bring in more people like them. If you institute enough diversity at the top, then this could theoretically trickle down.

Still, we have less diversity in senior levels of management, engineering, etc. (28% gender diversity in the tech industry broadly, but only 15% in engineering), so recommendations are often the worst way to ensure we have diversity. One of the best programs I ever saw was incentivizing the additional investment in recruiting diverse hires by creating programs that gave "free headcount," at least for the first year, to any manager bringing in under-represented minorities to the company.

We need to lower the risk of hiring so that managers have the incentive to take risks and go out of their comfort zone. There are other methodologies as well (hiring committees, removing "identifying" content from resumes, standardizing the questions, doing "buddy" interviews) that can normalize bias. This is rarely implemented systematically, but it dramatically impacts moving numbers.

Weekly: Many. First, standards (so we can actually start communicating and comparing across the complexity here). I mentioned before dynamic power modeling and more knobs from Silicon to systems to be able to tune performance and power dynamically (ideally with AI and correlation to workload placement), heat reuse methodologies (powering cities from servers), and exposure to end users for more dynamic choice (for example, do you want to watch this in high-definition and take six times the power, or is the standard definition "good enough" given you are watching on mobile).

Fundamentally, service owners usually try to compete with higher performance and higher quality. Still, suppose we add the lens of sustainability. In that case, we may actually see consumer behavior change (because they may not want "the best" if it is worse for the earth, and there is no way vendors will be comfortable making those investments until they know the market wants it).

DCK: One last question, what are you the most excited to see or do at DCW?

Weekly: This is my first data center world! In general, I always want to hear the range of pain points and solutions folks are experimenting with so we can engage in understanding where standardization can help drive solutions.

We’re excited to hear from Weekly and our other excellent speakers at this year’s AFCOM Data Center World conference. See you in Austin!

About the Author(s)

Bill Kleyman

Bill Kleyman has more than 15 years of experience in enterprise technology. He also enjoys writing, blogging, and educating colleagues about tech. His published and referenced work can be found on Data Center Knowledge, AFCOM, ITPro Today, InformationWeek, NetworkComputing, TechTarget, DarkReading, Forbes, CBS Interactive, Slashdot, and more.

Subscribe to the Data Center Knowledge Newsletter
Get analysis and expert insight on the latest in data center business and technology delivered to your inbox daily.

You May Also Like