Just one wasted byte per entry costs Cloudflare more than 250GB of fleet memory, which is a consequence of Big Pineapple holding more than “250 billion cached DNS entries at any given time.” The first of the five changes, replacing Rust's growable Vec and String containers with fixed-size boxed slices, saved over 15TB on its own by dropping capacity fields the data never uses once cached.
Further changes collapsed the three record lists in every DNS response into one buffer indexed by 2-byte offsets, and dropped owner names that duplicate the queried domain, rebuilding them at read time instead. The last change stores record data as length-prefixed raw wire-format bytes, ending an arrangement in which a 4-byte A record took up the same 144 bytes as the largest record type Cloudflare caches, the rarely seen NAPTR.
Packing record data contiguously also improved CPU cache locality, meaning that most record types now copy straight from the stored buffer into outgoing responses instead of being re-serialized field by field.
According to Cloudflare’s benchmarks in the technical blog, insert throughput is now 893,000 entries per second, up from 625,000, with lookup latency down from 828 to 670 nanoseconds. Over the rollout, which ran from mid-May to early July, p99 resident memory per instance fell from 9.3GB to 5.3GB.
Meta fights soaring hardware costs by reusing old DDR4 server memory in new DDR5-only servers
DDR5 scalper bots now outnumber shoppers 10 to 1 — automated scraping hits listings every 6.5 seconds
Marvell VP pushes for DDR4 recycling for use in CXL memory, amid the worst DRAM shortage in years — company introduces three-tier AI memory infrastructure
The Gen 13 servers, which Cloudflare calls its most powerful yet, each carry 768GB of DDR5-6400. The company settled on that config back in March after pricing out a 1,152GB option and rejecting it partly because of high memory prices.
This DNS cache work is the second large memory-reclamation project Cloudflare has completed in the last year, following the FL2 rewrite of its request-handling layer in Rust last September. Efficiency has been driving the company’s hardware picks for years, including the switch to AMD's 96-core EPYC 9684X for its Gen 12 servers .
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://www.tomshardware.com/tech-industry/big-tech/SPONSORED_LINK_URL
- https://www.tomshardware.com/tech-industry/big-tech/cloudflare-frees-100tb-of-ram-by-shrinking-dns-cache-entries#main
- https://www.tomshardware.com/membership
- X busts 200,000-strong Chinese bot farm, including accounts making claims about AI data centers and electricity — suspect accounts posted claims about pricing a
- EPA faces lawsuit over claims it fast-tracked approval of toxic 'data center chemicals' — exposure to photoacid generators used for semiconductor manufacturing
- Hot Chips 2026: OpenAI's Jalapeño AI ASIC unpacked — accelerator developed using AI achieves efficiency and throughput gains against power-hungry Blackwell
- NVIDIA NVLink Fusion Expands With NVHBM Custom High-Bandwidth Memory
- Google clamps down on Android app RAM usage amid AI memory crisis — developers have until February 2027 to adapt to new memory-optimizing rules
Informational only. No financial advice. Do your own research.