Saving 100 terabytes of memory by optimizing 1.1.1.1's DNS cache

Let's be honest: most "optimization" blog posts these days are thinly veiled marketing fluff that could be summed up as "we turned on compression and things got faster." This Cloudflare piece is the rare exception—a genuinely technical deep-dive that shows what happens when engineers obsess over every byte in a hot path. 100 terabytes of memory isn't a modest win; it's the kind of savings that lets you either shrink your fleet or absorb more traffic without buying new boxes. And they did it by attacking the DNS cache, the thing that sits in the critical path of literally every internet request that touches their resolver. The broader lesson here is one that open source folks have been preaching for decades: efficiency is a feature, not a footnote. We've gotten so used to the "just scale horizontally" mantra that we forget the cheapest server is the one you don't have to buy. Cloudflare isn't a scrappy startup—they have the resources to just add RAM and move on. But they chose to optimize instead, and that discipline is what separates good infrastructure shops from great ones. The fact that they're sharing this knowledge publicly, with real numbers and real tradeoff analysis, is exactly the kind of contribution that keeps the ecosystem healthy. What's particularly interesting is the tension they had to navigate between memory footprint and CPU cost. You can always compress harder, but at some point the decompression overhead eats your latency budget. The sweet spot they found—whatever it ends up being in the full post—is the kind of engineering judgment that doesn't come from a textbook. It comes from profiling, iterating, and actually understanding your workload. That's the stuff that doesn't get enough credit in a world obsessed with AI hype and new frameworks. The Hacker News thread (221 comments and counting) is worth a skim too. You'll see the usual armchair experts arguing about hash functions and cache eviction policies, but there's also real signal in there from folks who've run similar systems. That's the beauty of the open web: a blog post like this becomes a distributed code review, and everyone gets smarter as a result. Cloudflare continues to be one of the few big tech companies that treats its engineering blog as a genuine educational resource rather than a recruiting brochure. Long may it continue. Source: https://blog.cloudflare.com/dns-cache-memory-optimization-1111/