Mittagsseminar (in cooperation with A. Steger, D. Steurer and B. Sudakov)

Date and Time: Tuesday, February 07, 2017, 12:15 pm

Duration: 30 minutes

Location: OAT S15/S16/S17

Speaker: Dan Alistarh

The Power of Choice in Priority Scheduling

We consider the following random process, implementing a relaxed priority queue: we are given n queues, into which increasingly labeled elements are inserted uniformly at random. To remove an element, we pick two random queues, and remove the element of lower label (higher priority) among the two. The cost of a removal is the rank of the removed label among labels still present in any of the queues, that is, the distance from the optimal choice at step. Variants of this strategy are ubiquitous in practical implementations of concurrent priority queues; yet, no guarantees are known for this strategy.

This talk will give a solution to this question, showing that the expected cost of a removal is O( n ), while the expected worst-case cost is O( n log n ), irrespective of the number of steps for which the process runs. These bounds are asymptotically tight. The argument is based on a new connection between heavily loaded balls-into-bins processes and priority scheduling, and extends to the case where insertions are biased toward some of the queues, and removals may alternate between one and two random choices. This extension inspires a new concurrent priority queue implementation, which improves upon the state of the art in terms of throughput by up to 50%.

Joint work with Justin Kopinsky (MIT), Jerry Li (MIT), and Giorgi Nadiradze (ETH Zurich).

Upcoming talks     |     All previous talks     |     Talks by speaker     |     Upcoming talks in iCal format (beta version!)

Previous talks by year:   2024  2023  2022  2021  2020  2019  2018  2017  2016  2015  2014  2013  2012  2011  2010  2009  2008  2007  2006  2005  2004  2003  2002  2001  2000  1999  1998  1997  1996  

Information for students and suggested topics for student talks

Automatic MiSe System Software Version 1.4803M   |   admin login