There is a http://www.cs.utexas.edu/department/web

event/utcs/events/cgi/show_schedule.cgi?person=PetrosManiatis for this even

t.

Speaker Name/Affiliation: Petros Maniatis/Intel Research

Date/Time: February 5 2007 3:30 p.m.-4:30 p.m.

Location: ACES 2.4

02

Host: Lorenzo Alvisi

Talk Title: Distributed Systems as D

atabases: Specification Execution Debugging

Talk Abstract:
Deve

loping deploying and monitoring the execution of distributed
algorithm

s is a tough job whether in the context of routing
protocols and group
communication within general-purpose overlays
or deeply within focus

ed distributed applications. In P2 a joint
research project at Intel

Research Berkeley and UC Berkeley we start
with the premise that it is
desirable to trade off some design freedom
for a more manageable cycle
of design deployment and testing. Specifically we treat distributed sy

stems design at a high level as a sequence of asynchronous read-modify-wr

ite transformations of global
ground state --physical connectivity ava

ilable resources policy constraints past history -- much like how query p

rocessors regard traditional database problems. P2 translates such pseudoc

ode-level system specifications into software dataflow graphs made of small
interlinked reusable components which it then executes to run the speci

fied system. By managing the level of specification -- at pseudocode level
at dataflow graph level at traditional programming language level -- a p

rogrammer can range from a rough-and-ready rapid prototyping experience to

an increasingly tuned distributed application.

In this talk I will

present an overview of the architecture
design and implementation of

P2. I will then focus on the exciting
implications of a highly observab

le run time system for distributed
debugging on-line global invariant c

hecking and causality tracking.
Finally I will briefly describe promis

ing preliminary results on
specifying global optimization and adaptation
tasks as P2 queries
themselves as well as our current directions towar

ds automatic
parallelization and causally consistent execution.

S

peaker Bio:
Petros Maniatis is a Senior Researcher at Intel Research Be

rkeley. He holds Masters and Ph.D. degrees in Computer Science from Stanfo

rd
University. His research focuses on distributed systems fault
to

lerance and security.