Firmament: heterogeneity-aware task scheduling for warehouse-scale computers

Speaker: Malte Schwarzkopf , University of Cambridge Computer Laboratory

Date: Tuesday, November 12, 2013

Time: 4:00 PM to 5:00 PM Note: all times are in the Eastern Time Zone

Public: Yes

Location: 32-G882 (Hewlett Room)

Event Type:

Room Description:

Host: Frans Kaashoek

Contact: Frans Kaashoek, kaashoek@csail.mit.edu

Relevant URL:

Speaker URL: None

Speaker Photo:
None

Reminders to: seminars@csail.mit.edu, pdos@csail.mit.edu, pdos@pdos.csail.mit.edu, seminars@csail.mit.edu

Reminder Subject: TALK: Firmament: heterogeneity-aware task scheduling for warehouse-scale computers

With many-core machines populating today's warehouse-scale data centers
and many choices for partitioning "big data" computations across and
within machines, scheduling tasks in a large distributed data processing
system is an increasingly complex problem. Sharing of machine resources
and micro-architectural differences between machine generations also
introduce subtle yet severe interference effects across co-located jobs,
necessitating careful placement decisions.

In this talk, I will introduce the Firmament scheduler for task-parallel
data-flow computing. In particular, I will discuss how Firmament extends
Quincy's min-cost max-flow optimization approach in order to make global
resource-aware scheduling decisions in the presence of heterogeneity,
and present some early performance results. I will also discuss how
Firmament relates to a more broadly scoped research effort currently
under-way at Cambridge: DIOS, a re-visit of the distributed operating
system in the context of modern warehouse-scale data centers.

----

Malte Schwarzkopf is a final-year PhD student at the University of
Cambridge Computer Laboratory. His interests are in the areas of
distributed systems and operating systems, and especially where they
overlap. After being involved in the CIEL project (at Cambridge) and
working on the Omega cluster scheduler (at Google), he is currently
leading two related projects at the Computer Laboratory: the Firmament
heterogeneity-aware scheduler and DIOS, a distributed data-flow
operating system.

Research Areas:

Impact Areas:

This event is not part of a series.

Created by Frans Kaashoek Email at Wednesday, November 06, 2013 at 11:36 AM.