This paper discusses IO virtualization concepts allowing VMs to share physical IO devices safely and efficiently. Topics include emulated devices, paravirtualized drivers, direct device assignment (passthrough), and hypervisor support needed for isolating and multiplexing IO hardware.
Examines Facebook’s use of Memcache, a distributed, scalable in-memory key-value store caching data to reduce backend load and improve performance. Discusses architecture, consistency challenges, caching policies, failure handling, and optimizations for high throughput and low latency.
Haystack is a cloud storage system designed specifically for Facebook’s photo data. It reduces metadata overhead by storing multiple photos in large files, enabling low latency, high throughput, fault tolerance, and cost-effectiveness, overcoming limitations of traditional filesystems for large-scale media storage.
This document explains techniques for VM checkpointing and cloning to efficiently save and replicate VM states. It covers incremental checkpoints, compression, and how checkpointing supports VM migration, fault tolerance, and rapid provisioning for cloud environments.
Dynamo is a highly available, scalable distributed key-value store designed for fault tolerance and elasticity. It uses consistent hashing for data partitioning, vector clocks for conflict resolution, and quorum protocols to balance availability and consistency, making it ideal for services like shopping carts where eventual consistency suffices.