TAO: Facebook's Distributed Data Store for the Social Graph

By Nathan Grasso Bronson, Zach Amsden, G. Cabrera, Prasad Chakka, Peter Dimov, Hui Ding, J. Ferris, A. Giardullo, Sachin Kulkarni, Harry C. Li and 5 more. USENIX Annual Technical Conference, 2013

578 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/39ac2e0fc4ec63753306f99e71e0f38133e58ead

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Characterizing, Modeling, and Benchmarking RocksDB Key-Value Workloads at Facebook (2020, unknown): “On one hand, most read requests are responded to by the upper cache tiers [5, 7].”
Navigating the maze of graph analytics frameworks using massive graph datasets (2014, background): “There has been a lot of interest in creating, storing and processing large graph data [35, 10].”
LinkBench: a database benchmark based on the Facebook social graph (2013, background): “Facebook’s memcached and TAO cache clusters [2, 3] provide a caching layer”
PowerLyra: Differentiated Graph Computation and Partitioning on Skewed Graphs (2019, background): “For example, industrial-scale graphs like those from Google [47] and Facebook [6] are unlikely to be fit within a single machine, making distributed graph computation indispensable to process them in a timely manner.”
SoftSKU: Optimizing Server Architectures for Microservice Diversity @Scale (2019, background): “Prior work has reported that key-value stores, like Cache1 and Cache2, are typically memory intensive [16].”
Inside the Social Network's (Datacenter) Network (2015, background): “[ 15] The stability of these trafﬁc patterns bears special mention.”
Twine: A Unified Cluster Management System for Shared Infrastructure (2020, methodology): “For example, when migrating TAO [7], our social graph cache, from big machines to small machines, CacheLib [5] transparently provided tiered memory to improve cache hit ratio and reduce load on the external database by ≈30%.”
Accelerometer: Understanding Acceleration Opportunities for Data Center Overheads at Hyperscale (2020, background): “Since key-value stores are typically memory intensive [34], the low memory IPC (Fig.”
EmptyHeaded: A Relational Engine for Graph Processing (2015, methodology): “ process cold data from disk. We hope to create a system that spans the complete life cycle of data in the future. Dynamic Graphs. Finally, graphs which evolve over time occur in real-life workloads [3,12,42]. We hope to extend our system provide low latency mutations and ecient aggregates over chan”
FaRM: Fast Remote Memory (2014, unknown): “We also implemented a version of Facebook's Tao graph store [11] using FaRM.”
GraphJet: Real-Time Content Recommendations at Twitter (2016, background): “, [30, 5, 40]), we attempt to go beyond just describing a single system to comparing several generations of systems with similar goals.”
Rosetta: Large Scale System for Text Detection and Recognition in Images (2018, background): “haracters given each cropped word region from the image. (4) The extracted textual information along with the location of the text in the image is stored in TAO, Facebook’s distributed graph database [9] (step 6 in Figure 5). (5) Downstream applications such as Search can access the extracted textua”