Search posts, papers, and topics
Cloudflare reduced the memory footprint of its Pingora Backend Router by re‑examining the consistent‑hashing implementation in the pingora‑ketama library. By increasing the number of virtual hash points per server from the default 1 to the standard 160 (and applying weighted hashing based on disk capacity), they cut the per‑node overhead enough to reclaim >100 TB of RAM across the fleet. The post…
Pirate Face offers a decentralized, torrent‑based layer that mirrors open LLM models from Hugging Face, providing checksum‑verified downloads and a drop‑in API so pipelines stay functional even if the original host disappears.
The book *Warez* documents the underground piracy scene, tracing its infrastructure from early BBS networks to private FTP “topsites” and detailing its competitive, aesthetic culture. It argues that the scene’s drive is ranking and one‑upmanship rather than purely ideological motives.
Uber developed a context-aware mechanism to prevent retry storms in deep microservice dependency chains. It introduces "error ownership" where services claim errors they originate and unclaim errors they propagate, allowing upstream callers to make informed retry decisions and avoid amplifying load on already struggling services.
Agora treats a Git repository as a shared, append‑only memory for autonomous research agents, recording each claim as an immutable commit in a DAG. In a 12‑day run with 13 language‑model workers it generated 1,703 reproducible contributions and closed 62 % of the gap to a trained GPT‑2‑124M, demonstrating that shared research state can accelerate discovery.
The author built a Git server on object storage (objgit) and found standard Git packfiles inefficient due to random‑access and round‑trip latency. By designing a columnar, object‑storage‑native packfile format that can be accessed via HTTP range requests, they achieved acceptable performance for large repositories without client‑side changes.
Rheinmetall has open‑sourced onboardapi 9.10.0, a C++ middleware that uses DDS (XTypes, XCDR2) for low‑latency, backward‑compatible data exchange between sensors and software. The library includes Java, C#/.NET, and Python wrappers and comprehensive docs.
SiliconBench benchmarks nine Apple‑Silicon LLM serving engines on unified‑memory desktops, measuring speed, memory usage, and output fidelity across Qwen3, Qwen3.5, and Gemma‑4 models. It finds vllm‑metal doubles throughput at modest concurrency, memory budgets often fail to preserve headroom, and only three stacks satisfy all fidelity and model‑coverage requirements, with tensor‑parallel scaling…
Production AI agents often fail silently due to issues beyond model hallucination, such as state inconsistencies. This talk outlines principles for building reliable agent harnesses, focusing on explicit state ownership, ordered mutations, and verifiable actions, using real-world case studies.
Atlassian replaced its decade‑old gostatsd‑based metrics pipeline with a fully OpenTelemetry‑based stack by keeping the StatsD‑UDP contract on the client side and swapping in purpose‑built OTel Collector distributions for collection, ingest, aggregation, and forwarding. The migration was done incrementally, saved ~3.9% CPU per service, cut sidecar cost ~30% fleet‑wide, halved aggregation CPU, and…
Squalk is a SvelteKit‑based forum built on the Nostr protocol, implementing NIP‑29 groups and NIP‑7D threads. It can run in a single‑forum “simple” mode or a multi‑forum “full” mode, with chat sidebars, markdown resources, and optional server‑side rendering for SEO. Configuration is done entirely via `PUBLIC_` environment variables, and deployment scripts support both static hosting and Node SSR,…
Neki is PlanetScale’s sharding layer for vanilla PostgreSQL that presents a single Postgres endpoint while routing queries across a fleet of Postgres instances. It does this with a set of tightly‑coupled components—Router, Sidecar, PostgresManager, Admin, Operator, and etcd‑backed Data Topology—each handling a specific piece of the scaling, failover, and query‑planning puzzle.
Canva added a lightweight, local backpressure loop to its queue worker library that monitors per‑message success/failure, computes a backoff factor against a configurable failure‑rate set‑point, and throttles the worker’s concurrency. In two real incidents the mechanism kept failure rates under 2 % fleet‑wide, limited DLQ growth to a handful of messages, and maintained throughput without manual i…
NVIDIA’s DSX platform lets AI data‑centers shift workloads in response to grid signals, squeezing ~24% more token throughput (4 M→5 M tps) and ~23% better performance‑per‑watt on a fixed megawatt budget. The first production demo used Emerald AI’s Conductor to drop a 4 MW load to 3 MW in under a minute without interrupting high‑priority jobs. DSX MaxLPS reallocates headroom across HGX B200 server…
Meta open‑sourced Rebalancer, a library that lets engineers describe assignment problems (objects → bins) via a high‑level spec API, then solves them with either a MIP backend or a highly parallel local‑search engine. It handles millions of objects, solves ~40 M problems daily, and includes a debugging UI.
Pinterest reengineered its Manas search platform to replace memory‑heavy HNSW with scalar and product quantization, cutting index size by up to 93% while keeping recall above 70% and achieving 20‑30% cost savings. They also moved ANN storage to SSD using SPANN, gaining 3× query throughput over DiskANN with modest latency and recall impact.
Uber reworked M3DB's shard placement by grouping nodes into fixed-size subclusters, each owning a disjoint shard range, to bound the blast radius of node failures. A greedy shard‑move algorithm (O(S log S) + O(S×N)) handles scaling while preserving existing placement tooling.
Replica‑aware routing (public beta) lets ClickHouse Cloud users pin a query stream to a specific replica by sending a custom header (HTTP) or overriding the TLS SNI (native). The proxy (Envoy) hashes the tag and consistently forwards all requests with the same tag to that replica, giving read‑after‑write consistency for temporary tables, session objects, and warm replica caches. Stickiness is bes…
This guide details building a distributed sliding-window rate limiter using TypeScript and Redis. It leverages Redis sorted sets and an atomic Lua script to prevent race conditions and over-admission under high concurrency, supporting dual RPM/TPM quotas.
A real‑world post‑mortem of a high‑volume face verification service that moved from a naïve synchronous API to an async, layered pipeline (edge validation, preprocessing, decoupled detection/verification, decision engine) to achieve 8.5k rpm, p99 < 1.8 s, 30 % cost savings, and strict privacy controls.
Salesforce’s Data 360 provides a shared runtime that assembles the minimal, authorized slice of enterprise data (“Trusted Context”) for each AI‑agent turn. A six‑stage Agent Context Engine (Resolve, Plan, Reconcile, Govern, Compile, Learn) pulls data from structured, unstructured, and streaming sources across Salesforce, Snowflake, Databricks, etc., applies fine‑grained policy, and returns a toke…
Dropbox’s Riviera platform has been expanded from a preview service into a universal, composable content‑processing system that now handles hundreds of thousands of transformations per second across 300+ file types. The architecture separates orchestration from execution, adds a plugin model, and offers async public APIs so internal and external AI workloads can reuse cached transformations.
Google Cloud’s blog introduces Distributed GraphFlow (DGF), an open‑source Python library for building and scaling Graph Neural Networks (GNNs) on a Spanner‑backed digital twin of telecom networks. The post outlines the three‑layer architecture (digital twin on Spanner Graph, ML layer with DGF, AI agents) and highlights DGF’s high‑level API (5‑line example) and low‑level primitives, but provides…
Google’s DevFest Community Workshop introduced a “Workbench” format that emphasizes architectural mental models over copy‑paste code, guiding engineers to build long‑running, self‑evolving multi‑agent systems with the Agent Development Kit and Gemini Enterprise platforms. Attendees learned state‑separation, workflow pausing, and self‑patching pipelines, and the series will continue in five more c…
This article distinguishes between serialization (preventing concurrent access) and arbitration (permitting access and rejecting losers) in distributed concurrency control. It argues that arbitration should be responsible for correctness, as serialization guarantees are conditional and can fail silently in multi-process environments, leading to data corruption.
This Jepsen blog post announces a retrospective talk on 13 years of testing distributed systems, linking to slides and a video. The post itself contains no detailed lessons or technical content.
Datadog explains how scaling AI agents turns isolated executions into shared‑platform risk and outlines a systematic approach to model agent trajectories, monitor per‑dependency constraints, and enforce workload‑specific capacity policies. The result is proactive detection of bottlenecks and protection against noisy‑neighbor failures.
The piece shows that data‑role system design interviews evaluate how you turn vague requirements into a defensible architecture, not which tools you name, and it gives a concrete prep framework: clarify scope, quantify load, pick batch vs streaming with trade‑offs, and address failure handling. Candidates should rehearse explaining these decisions aloud with numbers rather than just drawing diagr…
API design is shifting from human‑centric, resource‑based REST toward machine‑friendly, capability‑oriented interfaces as AI agents become major consumers. This drives changes in rate limiting, just‑in‑time security, system‑wide reliability, and regulatory compliance.
