O-9: distarray global API -- array expressions with partially overlapping arrays (optional)

Open
#189 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
20/100
Issue type
Feature
Clarity
Needs clarification
Activity status
Stale
Tech stack
python
Domain
distributed-systems

Research direction

Start by clarifying the semantics for expressions combining partially overlapping distributed arrays, including result distribution and which nodes perform computation. Compare the desired behavior with NumPy functionality, then evaluate the communication-versus-computation tradeoffs and possible user controls such as decorators or context managers. Done means the ambiguities are resolved and an implementation direction is defined.

Written by the indexing model from the issue text.

Description

SBIR-year-2

Objective: The goal of this sub-task is to define the semantics for distributed array expressions with arrays that are not identically distributed.

Relevance: Supporting array expressions when the component arrays are not identically distributed will allow ODIN to more closely align with NumPy’s functionality, will improve overall usability, and will more closely conform to what is expected for users coming from a NumPy background (i.e. the principle of least surprise).

Description: When partially overlapping distributed arrays are combined in an expression whose result is another distributed array, certain ambiguities arise: how should the result array be distributed? What nodes are responsible for the computation? In which circumstances is communication more costly than computation, and what heuristics can be devised to help guide these decisions at runtime?
Given these ambiguities, allowing end-user control in some form is a possible option to pursue, whether it be via decorators, context managers, or other mechanisms provided by Python control flow structures.

Dominant language
Python
Stars
5
Forks
1
No merged PRs in 30d

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from enthought/distarray

All issues in enthought/distarray

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.