Skip to content

Latest commit

 

History

History
18 lines (17 loc) · 2.17 KB

README.md

File metadata and controls

18 lines (17 loc) · 2.17 KB
Raw

Package documentation

  • budget: various implementations for the privacy budget
  • task.py: the Task specification
  • cache: deterministic / probabilistic caches for storing results. Implemented as Redis key-value stores. There are 'Mock' versions for both where the key value stores are implemented as simple in-memory dictionaries. (redis version pending for probabilistic)
  • planner: max_cuts / min_cuts / ILP - our three versions of the planner
  • budget_accountant.py: Redis key-value store for storing the budgets of blocks. There is a 'Mock' version where the key value store is implemented as a simple in-memory dictionary.
  • psql: an API for storing block data to TimeScaleDB using and running SQL queries using a PSQL client. There is a 'Mock' version that stores block data in a in-memory dictionary as "histograms" instead, and runs queries using tensor operations.
  • server_blocks.py: a Blocks Server listening to a socket for new requests for adding block data (both in TimeScaleDB and in the budget_accountant).
  • server_tasks.py: a Tasks Server listening to a socket for new requests for running a query.
  • client_blocks.py: API for sending requests to the Blocks Server
  • server_blocks.py: API for sending requests to the Tasks Server
  • query_processor.py: Finds a DP plan for the query if possible, runs it, consumes budget if necessary and stores metadata.
  • executor.py: executes the DP plan of a query using the caches and the PSQL module.
  • simulator: a simulation of the execution of turbo implemented using Simpy. It generates a workload and data blocks given the configuration in turbo.json. It bypasses the blocks/tasks servers API and directly uses the rest of the package modules to execute queries and store data blocks and block budgets. All experiments in the evaluation use this module to simulate requests on Turbo.
  • run_simulation.py: entrypoint for running turbo in a simulation.
  • turbo.json: configuration file to setup the turbo execution. Contains configuration for the simulation as well. To run everything using the mock modules set the flag "mock": true.