TL;DR
This paper reports on a panel discussion at the 14th International Parallel and Distributed Processing Symposium (IPDPS 2000) where leading computer scientists identified the ten most influential parallel and distributed processing concepts of the last millennium. The findings, based on panelist presentations and an audience vote, offer a historical perspective on the foundational ideas that shaped modern computing.
What Problem Does This Paper Address?
As the millennium turned, there was a desire to reflect on and identify the most impactful technological advancements. This paper addresses the challenge of distilling the vast landscape of parallel and distributed processing concepts into a definitive "top ten" list that represents the most influential ideas of the past thousand years. The problem lies in the subjective nature of such a selection, the broad scope of "concepts," and the need to represent diverse perspectives within the field.
Methods
The paper details a panel discussion format designed to gather expert opinions and then solicit broader input.
- Panelist Selection: Eight senior computer scientists and engineers were chosen to represent a variety of perspectives and technical areas.
- Panel Presentations: Each panelist presented their "top ten list" of influential concepts, with a 10-minute time limit. The ordering was arbitrary, and panelists could present fewer than ten items.
- Open Discussion: An open discussion followed the presentations, allowing for audience participation and questions.
- Audience Survey: An audience ballot was distributed for participants (including panelists) to vote on the top ten concepts in arbitrary order.
- Analysis: The results of the audience vote were compiled to determine the final top ten list.
Datasets
Not discussed in the paper. The "data" here is the collective opinion of the panelists and audience members, gathered through presentations and a ballot.
Key Insights and Ideas
- Defining "Influence" and "Concept": The discussion highlighted the ambiguity of these terms, with panelists debating whether concepts should be broad and abstract (like "divide and conquer") or more specific and technical.
- Subjectivity vs. Consensus: While individual panelists had diverse lists, the audience vote aimed to find a consensus on the most impactful ideas.
- Historical Context: Many panelists acknowledged that some concepts predated the millennium being considered, prompting a discussion on how to interpret "influence" over longer timescales.
- The Role of the Market: One panelist emphasized the market's influence on the adoption and perceived importance of parallel computing ideas, suggesting that good ideas might not gain traction if they don't align with market demands.
- Broadening the Definition of Parallelism: Some panelists, like Larry Snyder, proposed looking beyond traditional computing paradigms to find parallels in everyday life and natural systems (e.g., the Parthenon as a parallel architecture, the solar system's gravitational interactions).
- The "Chicken or the Egg" Dilemma: Ken Kennedy raised the question of whether concepts became familiar because they were influential, or influential because they were familiar.
Key Results
The paper culminates in Table 12, which presents the top ten most voted concepts from the audience survey:
- Amdahl's Law and scalability (96 votes)
- Arpanet and Internet (95 votes)
- Pipelining (79 votes)
- Divide and conquer (63 votes)
- Multiprogramming (53 votes)
- Synchronization (including semaphores) (50 votes)
- Load balancing (42 votes)
- Message passing and packet switching (42 votes)
- Cluster computing (40 votes)
- Multithreaded (lightweight) program execution (39 votes)
Limitations and Future Work
- Subjectivity: The inherent subjectivity of selecting "influential" concepts remains a limitation, even with an audience vote. The results reflect the opinions of the participants at that specific symposium.
- Millennium Definition: The paper touches upon the challenge of defining the "past millennium" and how concepts might have evolved or been discovered at different points.
- Scope of Concepts: The broadness of "concepts" led to varied interpretations, from fundamental algorithms to networking infrastructure.
- Future Work: The paper doesn't explicitly outline future work but implies that such discussions and historical reflections are valuable for understanding the trajectory of the field. The results themselves serve as a benchmark for future historical analyses of parallel and distributed computing.
Summary updated