Efficent Window Aggregation in Decentralized Networks

Desis

Desis is a hierarchical system that can be deployed into decentralized networks. Desis support tumbling windows, sliding windows, session windows, and user-defined windows.

Deco

Deco is also a decentralized approach that is based on Desis. Deco can move calcuations of count-based windows from centers to the local nodes.

Approach Introduction:

• Desis
  • decentralized aggregation and share results between windows
• Deco
  • decentralized aggregation and supports count-based windows
• Scotty
  • implemented by the Scotty
• Disco
  • implemented by the Disco
• DesisCen(DeCen)
  • centralized aggregation and can not perform incremental aggregation.
• DesisIC(DeBuket)
  • implemented based on Desis and only can perform incremental aggregation
• DesisSW(DeSW)
  • implemented based on Desis and can partial results between windows have different window types
• DesisMultipleKeys
  • the same as Desis and supports windows with different distinct keys
• DesisQuantile
  • the same as Desis and supports different quantile functions
• FigureScripts
  • the python scripts that draw experiments plots

Installation

• Requirements: Java 8+
• Install
  1. Download DESengine
  2. Compile Project
  3. Set WINDOWS = true in Configuration.java
  4. Run OverallMainDriverTest.java

Input

• Desis

  • Node Id:
    • The id of the node that is deployed with Desis.
  • Query Number:
    • How many queries are processed simultaneously.
  • Query Modes:
    • The query mode is to choose the query pattern that is set into query generation file.
  • Generator Thread Number:
    • How many generator threads are initialized. One thread can produce at least 10 million tuples/s.

• Deco

  • Node Id:
    • The id of the node that is deployed with Desis.
  • Local Nodes:
    • How many local nodes.
  • Query Modes:
    • The query mode is to choose the query pattern that is set into query generation file.
  • Generator Thread Number:
    • How many generator threads are initialized. One thread can produce at least 10 million tuples/s.

Experimental Data

https://zenodo.org/record/8120787/files/DEBS-Data.tar.gz?download=1