SEDA: an architecture for well-conditioned, scalable internet services

By Matt Welsh, David E. Culler, Eric A. Brewer. Symposium on Operating Systems Principles, 2001

1,064 citations recorded by the source corpus.

Subfields: Computer Networks and Communications

View Paper: https://www.semanticscholar.org/paper/e3966d51c793d6877a9aa0d690cb52011f75ddd7

How researchers see this paper

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

In Computer Networks and Communications

Citations in the corpus

Pond: The OceanStore Prototype (2003, methodology): “We built Pond in Java, atop the Staged Event-Driven Architecture (SEDA) [39], since prior research indicates that event-driven servers behave more gracefully under high load than traditional threading mechanisms [39].”
Clipper: A Low-Latency Online Prediction Serving System (2016, background): “The division of functionality into vertical stages introduced by [61] is similar to the division of Clipper’s architecture into independent layers.”
Labels and event processes in the asbestos operating system (2005, background): “Many efficient servers [20, 33, 43, 45] use an implementation pattern that suggests a solution to this problem.”
Corey: An Operating System for Many Cores (2008, positive): “, running stages of a Web server like SEDA [24] on different cores).”
Pip: Detecting the Unexpected in Distributed Systems (2006, neutral): “For example, we believe that SEDA [27] and RPC platforms like CORBA could generate message and ask events and could propagate path IDs.”
A method for transparent admission control and request scheduling in e-commerce web sites (2004, background): “SEDA decomposes Internet services into multiple stages, each one of which can perform admission control.”
Low latency RNN inference with cellular batching (2018, background): “Compared with traditional pipelining (in the context of hardware pipeline and software pipelining, e.g. SEDA[45]), our system is different in two aspects: 1) Traditional pipelining involves multiple processing elements each of which is sequential and operates independently of each other.”
Efficient Byzantine Fault-Tolerance (2013, methodology): “To achieve this goal, we built a scalable event-driven I/O architecture (which can be seen as a simpler version of SEDA [47]) and implemented an adaptive batching algorithm and window congestion control similar to the one used in PBFT (the algorithm can run a pre-configured maximum number of paralle”
TensorFlow-Serving: Flexible, High-Performance ML Serving (2017, background): “Lastly, there is of course a great deal of literature on web serving and other non-ML serving scenarios, e.g. [10–13], which has informed our work.”
State Machine Replication for the Masses with BFT-SMART (2014, background): “Figure 3 presents the main architecture with the threads used for staged message processing [34] of the protocol implementation.”
Renew (2022, methodology): “To scale the processing to many cores, Agora employs a carefully designed threading model, borrowed from web server design [26, 22], and applies a series of non-trivial cache-aware optimizations to cope with the memory bottleneck.”
Dynamo: amazon's highly available key-value store (2007, methodology): “The request coordination component is built on top of an eventdriven messaging substrate where the message processing pipeline is split into multiple stages similar to the SEDA architecture [24].”