Search Results for "replication"

EPLB is DeepSeek’s open implementation of a load balancing algorithm designed for expert parallelism (EP) settings in MoE architectures. In EP, different “experts” are mapped to different GPUs or nodes, so load imbalance becomes a performance bottleneck if certain experts are invoked much more often. EPLB solves this by duplicating heavily used experts (redundancy) and then placing those duplicates across GPUs to even out computational load. It uses policies like hierarchical load balancing...

This project provides an open-source setup for IPFS Kubo (public and private pinning) combined with IPFS Cluster replication, designed for reliable, production-ready decentralized storage. It is suitable for developers, infrastructure operators, and Web3 projects that need predictable IPFS availability without maintaining complex node infrastructure. The system supports multiple retention plans for IPFS Kubo pinning, allowing users to choose how long data remains pinned, with transparent pay-per-GB prepaid pricing. ...

...In addition to standard NAS features like file sharing via NFS, Samba, SFTP and AFP, advanced features such as online volume management, CoW Snapshots, asynchronous replication, compression, and bitrot protection are supported based on BTRFS. Rockstor also provides apps like ownCloud, Syncthing, OpenVPN, and Plex to name a few. Apps or "Rock-ons" are powered by a Docker-based application hosting framework. And new ones can be simply added. These Rock-ons, combined with advanced NAS features, turn Rockstor into a private cloud storage solution accessible from anywhere, giving users complete control of cost, ownership, privacy and data security. ...

Spilo is a Docker-based HA PostgreSQL cluster built on Patroni and heavily optimized for Kubernetes environments. It includes components for failover, streaming replication, backups, and connection pooling. Spilo is used in production by Zalando and is designed to provide a resilient, self-healing Postgres cluster with minimal manual intervention.

This system is designed to synchronize objects over network using a 1 server : n clients topology. In contrast to Object Replication system used in Zoidcom or RakNet this solution is developed to be very scalable using multiple threads/cpus=> MMO ready.

A python daemon to watch a directory for kernel’s inotify events and then execute an rsync command to synchronize two different directories (local or remote). A poor man’s mirroring or an alternative (not so) real data replication mechanism

SF2 is a generic framework for constructing cloud computing platform. The framework provides snapshot-based replication, virtual machine placement on multiple sites, and gossip-based physical machine discovery.

A universal monitor written in Python. It is a basic monitor framework with pluggable modules for monitoring things like log files, MySQL replication status, and other things that software like Zenoss do not monitor very well.