POSITION: Postdoctoral fellowship

TITLE: Cost-Aware Resource Management in Elastic Cloud Computing Infrastructures

PLACE: INRIA, Grenoble, France

DEADLINE TO APPLY: The earlier the better.  Best to first
send CV by email and in parallel, ask letters to be sent.

DATE AND DURATION: Starting date and duration are flexible. Preferably
starting in September 2011.  Duration can be up to 24 months.

SUPERVISOR: Derrick Kondo

CONTACT INFORMATION: derrick.kondo :: inria.fr
http://mescal.imag.fr/membres/derrick.kondo/

KEYWORDS: Cloud Computing, Resource Management, Performance Modeling,
Parallel and Distributed Computing

TOPIC:

Current users of Cloud Computing platforms have an enormous number of
tiered choices when it comes to server selection.  For instance, Amazon
offers 10 types of server sizes that differ in memory amounts, network
bandwidth, and CPU speeds and cores, and it offers 3 types of storage that
differ in reliability and transmission speeds. When one looks at resource
types of other Cloud providers, the number of combinations of server types
grows exponentially.

Given an application such as a web service or batch job, this begs the
question of whether it should be run on X servers of type Y in Cloud
Provider Z, or A servers of type B in Cloud Provider C, or some
combination? The problem is important and different from traditional
distributed computing environments such as Grids because the number of
servers are virtually unlimited, the choice can be dynamically adjusted
via cloud virtual machine technologies, and the resources costs among
server types and Cloud providers differs tremendously and is explicit.

For the postdoctoral position, we will use a statistical approach to conduct the
performance modeling of applications on real Cloud Computing platforms,
namely, Amazon's Elastic Compute Cloud.

In particular, we will focus on the following issues:

(1) the development of macro and micro-benchmarks of storage, memory, CPU,
and network resources in Clouds

(2) the construction of application signatures for predicting their
resource usage

(3) the evaluation of techniques that partially sample applications in
order to predict performance with different server types of different
vendors.

We will apply this approach using common web services (with frameworks
such as RUBIS) and real batch jobs from production Desktop Grids.
Experiments will be conducted on Amazon Inc's Elastic Compute Cloud.

This work will help users achieve desired levels of performance with
minimal monetary costs in large-scale Cloud Computing infrastructures.

SKILLS AND KNOWLEDGE: Experience with statistics and probability, machine
learning, parallel and distributed computing, and performance modeling are
all desired.

TO APPLY:  Send CV along with 3 letters of recommendation to derrick.kondo
:: inria.fr