Tag
The article explores the viability of home-based GPU networks as supplements to centralized AI data centers, addressing community opposition and the need for distributed computing solutions.
The paper introduces Tsampi BFT, a leaderless one-round voting protocol with parameterized finality for achieving Byzantine fault-tolerant consensus in distributed systems.
This paper presents the modeling and verification of the Keeta consensus protocol, focusing on formal methods to validate its correctness in distributed systems.
TLA+ is a model checking tool that explores all state interleavings to find bugs in distributed systems; it helped improve safety in Depot Registry's garbage collector by identifying a missed bug through formal verification.
The TiDB team deliberately avoids using Multi-Agent systems internally, based on lessons learned from distributed systems and referencing Anthropic's related research, emphasizing that optimizing a single Agent might be more effective.
TideRL is a readiness-aware elastic RL system that improves training goodput for multi-turn agentic workloads via continuous task batching, resource-aware ref-actor pipelining, and elastic resource scaling, achieving up to 5.6x speedup over synchronous baselines and 33% over asynchronous baselines.
A tweet highlights a GitHub repository containing comprehensive system design interview notes based on Alex Xu's books, covering topics like rate limiting, consistent hashing, key-value stores, and more.
This paper presents DNSSE, a hybrid framework combining LLM-guided schedule prediction, symbolic constraint solving, and coverage-guided stochastic mutation for verifying distributed parallel AI programs. It detects 2.9x more concurrency bugs than the baseline and raises branch coverage from 68.6% to 91.6% on realistic benchmarks.
The paper presents a two-level algorithm for allocating a conserved capacity budget across locations and service classes, proving it conserves the budget and converges in one iteration, and evaluates it for defending CDNs under volumetric attack.
A tweet from @venkateshdotdev listing 10 challenging system design interview questions covering URL shorteners, scaling, consistency, rate limiting, fault tolerance, and handling traffic spikes.
Choral is a choreographic programming language for Java that lets developers write distributed systems in a single program, with the compiler generating endpoint code.
An in-depth technical article explaining the ABD algorithm, quorum replication, and why ABD does not solve consensus, with runnable Python examples.
celld is a self-hosted, distributed server that runs Cloudflare Workers and Durable Objects code unchanged, using your own S3 bucket for storage and coordination. It claims significantly lower costs and sub-millisecond warm latencies at scale, with data stored in SQLite/LTX format.
Explains the AT Protocol architecture through the lens of distributed systems engineering, covering the shift from SQL to NoSQL to stream processing, and how AT Protocol externalizes these services for a decentralized backend.
AWS engineer Zak van der Merwe shares insights from his 14 years building control planes for EC2 and DSQL, discussing the distributed systems challenges of running infrastructure at massive scale.
Deno's celld is an open-source daemon that runs Cloudflare Workers and Durable Objects on your own machines, using SQLite per object and an S3-compatible bucket for replication and coordination without a control plane.
This paper proposes the Agent Operating System (AOS), a vendor-neutral reference operating architecture for distributed agentic systems, covering governance, runtime coordination, and reliability. It aims to provide a stable framework for composing heterogeneous components into governable and observable agentic systems.
Aquaduck, an aspiring inference platform, is seeking beta testers for its multiplayer model serving network that turns desktop devices into a virtual cluster, with features like model sharding and a public inference network for idle compute.
The article argues that retrying messages in eventually consistent distributed systems treats normal state as failure, and instead suggests storing incoming data and processing when all prerequisites arrive, eliminating the need for retries and dead-letter queues.
A production engineer describes a race condition in multi-agent shared memory, caused by concurrent writes, and explains how switching to an append-only event log with projection resolved it while noting the trade-off in read-your-own-write latency.