The Scheduler is Very Powerful in Competitive Analysis of Distributed List Accessing

TL;DR

This paper explores the impact of adversarial schedulers on the competitive analysis of distributed list accessing algorithms, revealing that scheduling can significantly influence algorithm performance beyond traditional information limitations.

Contribution

It introduces new adversaries and bounds for distributed list accessing, highlighting the scheduler's powerful role in competitive analysis.

Findings

Adversarial scheduler effects can dominate traditional performance loss.

Established tight bounds on request sequence merges.

Distributed setting differs fundamentally from classical online models.

Abstract

This work is a continuation of efforts to define and understand competitive analysis of algorithms in a distributed shared memory setting, which is surprisingly different from the classical online setting. In fact, in a distributed shared memory setting, we find a counter-example to the theorem concerning classical randomized online algorithms which shows that, if there is a $c$-competitive randomized algorithm against an adaptive offline adversary, then there is a $c$-competitive deterministic algorithm [Ben-David, Borodin, Karp, Tardos, Wigderson, 1994]. In a distributed setting, there is additional lack of knowledge concerning what the other processes have done. There is also additional power for the adversary, having control of the scheduler which decides when each process is allowed to take steps. We consider the list accessing problem, which is a benchmark problem for sequential online algorithms. In the distributed version of this problem, each process has its own finite sequence of requests to a shared list. The scheduler arises as a major issue in its competitive analysis. We introduce two different adversaries, which differ in how they are allowed to schedule processes, and use them to perform competitive analysis of distributed list accessing. We prove tight upper and lower bounds on combinatorial properties of merges of the request sequences, which we use in the analysis. Our analysis shows that the effects of the adversarial scheduler can be quite significant, dominating the usual quality loss due to lack of information about the future.