Cooperative Scheduling for Cloud Computing · Bernabé Dorronsoro • Computer Science and PhD,...
Transcript of Cooperative Scheduling for Cloud Computing · Bernabé Dorronsoro • Computer Science and PhD,...
Cooperative Scheduling for Cloud Computing
Bernabé Dorronsoro · Juan J. Durillo
Bernabé Dorronsoro• Computer Science and PhD, University of Málaga, Spain
- Title: Design and Implementation of Cellular Genetic Algorithms for Complex Problems, 2007. !
• Supervised by: Prof. Enrique Alba !
• Post-DocUniversity of Luxembourg (Sept. 2007 - Aug. 2010) !
• Post-Doc SnT (Sept. 2010 - Aug. 2012) !
• AFR Post-Doc Univ. of Lille 1 (Sept. 2010 - Now) ���2
www.bernabe.dorronsoro.es
FNR-FWF Bilateral Projects
• From 2012 • All fields of research • Details
- Submitted to the organization where the main research will be done ‣ If Luxembourg -> Through CORE
‣ If Austria
‣ No deadline
‣ ~90K€ per year
‣ Success rate ~30%
!- Evaluated only in one country
- Funded by FWF and FNR
- Maximum 3 years
���3
Current Picture
���4
Cooperative Scheduling
for
Green Cloud Computing
Recent years have witnessed a change in the way computing is done by scientists and institutions. This
change has brought about a switching from local IT infrastructures to data centers built by companies
like Google, Amazon, etc. The number of these data centers is only expected to increase in the near
future due to the growing demand of computing services.
A capital problem for data centers is the enormous amount of power required for operating them which,
besides environmental issues, make up an important percentage of their operation cots. In such a
situation, different techniques are adopted in order to reduce the energy consumption. Example of these
techniques are: dynamic server configuration—powering off non-used nodes–, CPU frequency Scaling
—scaling down the frequency of CPU when no peak performance required –, or price and location
diversity– which takes advantage of fluctuating price of energy in different countries, locations to
migrate workload. In the context of Cloud computing, an example of dynamic server configuration is
consolidation: a well-known strategy that tries to group virtual machines (VM) into a reduced set of
physical machines (PM). To achieve this, the provider schedules VM into PM; in the following we refer
to this as provider schedule.
Nowadays, workflows have become a popular paradigm for building parallel applications in the
scientific domain. A workflow consists of a graph, where each node represents a computational task
that needs to be executed and arcs represents dependences between these tasks (e.g., data
transferences). One research challenge consists in mapping the workflow's tasks onto computational
resources in such a way that one or several criteria gets optimized. Examples of these criteria are
completion time, execution costs, energy consumption, or reliability, to mention a few. In the following
we will refer to this as workflow schedule.
The main problem in a real scenario is that both, the provider and the workflow schedules, are done
independently and their decisions might collide with each other's interests. An example of this may
may occur when both the user and provider tries to optimize for energy consumption. The workflow
schedule may determine a given number of VM of a given type is the solution minimizing the energy
consumption. This decision may however difficult the consolidation of VM onto PM, thus resulting on
user's goals provider's goals● Minimize make-span● Minimize cost● Minimize energy● Minimize risk
● Minimize operation costs
● Maximize revenue● Customers satisfaction
● Minimize energy● ...
scheduling
set of VM
makespan
cost
SLA(Service LevelAgreement)
virtual machinemigration
United Kingdomdata center
Austriandata center
Inet
The Idea
• Current approach - Not optimal
- User’s and provider’s interests are in conflict !
• What do we look for? - Best performance and price for users
- Best efficiency for provider !
• How? - Address the whole problem
- Coupling user and provider decisions ‣ Priorities
‣ Relaxed SLA
���5