Max-Planck-Institut für Informatik
max planck institut
mpii logo Minerva of the Max Planck Society

MPI-INF or MPI-SWS or Local Campus Event Calendar

<< Previous Entry Next Entry >> New Event Entry Edit this Entry Login to DB (to update, delete)
What and Who
Title:Building Fast and Consistent (Geo-)Replicated Systems: From Principles to Practice
Speaker:Cheng Li
coming from:Max Planck Institute for Software Systems
Speakers Bio:
Event Type:SWS Student Defense Talks - Thesis Defense
We use this to send out email in the morning.
Level:Public Audience
Date, Time and Location
Date:Monday, 30 May 2016
Duration:90 Minutes
Building:E1 5
Distributing data across replicas within a data center or across multiple
data centers plays an important role in building Internet-scale services
that provide a good user experience, namely low latency access and high
throughput. This approach often compromises on strong consistency
semantics, which helps maintain application-specific desired properties,
namely, state convergence and invariant preservation. To relieve such
inherent tension, in the past few years, many proposals have been designed
to allow programmers to selectively weaken consistency levels of certain
operations to avoid costly immediate coordination for concurrent user
requests. However, these fail to provide principles to guide programmers
to make a correct decision of assigning consistency levels to various
operations so that good performance is extracted while the system behavior
still complies with its specification.

The primary goal of this thesis work is to provide programmers with
principles and tools for building fast and consistent (geo-) replicated
systems by allowing programmers to think about various consistency levels
in the same framework. The first step we took was to propose RedBlue
consistency, which presents sufficient conditions that allow programmers
to safely separate weakly consistent operations from strongly consistent
ones in a coarse-grained manner. Second, to improve the practicality of
RedBlue consistency, we built SIEVE - a tool that explores both
Commutative Replicated Data Types and program analysis techniques to
assign proper consistency levels to different operations and to maximize
the weakly consistent operation space. Finally, we generalized the
tradeoff between consistency and performance and proposed Partial
Order-Restrictions consistency (or short, PoR consistency) - a generic
consistency definition that captures various consistency levels in terms
of visibility restrictions among pairs of operations and allows
programmers to tune the restrictions to obtain a fine-grained control of
their targeted consistency semantics.
Video Broadcast
Video Broadcast:YesTo Location:Kaiserslautern
To Building:G26To Room:111
Meeting ID:
Tags, Category, Keywords and additional notes
Attachments, File(s):
  • Maria-Louise Albrecht, 05/11/2016 10:26 AM -- Created document.