Low Carbon Virtual Private Clouds Fereydoun Farrahi Moghaddam, Mohamed Cheriet, Kim Khoa Nguyen Synchromedia Laboratory Ecole de technologie superieure, Montreal Presented By, Chidambara Nadig. 27 th November, 2012.

Abstract  Data center energy efficiency and carbon footprint reduction have attracted a great deal of attention across the world for some years now, and recently more than ever.  Live Virtual Machine (VM) migration is a prominent solution for achieving server consolidation in Local Area Network (LAN) environments.  With the introduction of live Wide Area Network (WAN) VM migration, however, the challenge of energy efficiency extends from a single data center to a network of data centers. Low Carbon Virtual Private Clouds 2

 In this paper, live migration of VMs within a WAN is used as a reallocation tool to minimize the overall carbon footprint of the network.  Simulation results show that using the proposed Genetic Algorithm (GA)-based method for live VM migration can significantly reduce the carbon footprint of a cloud network compared to the consolidation of individual data center servers.  WAN data center consolidation results show that an optimum solution for carbon reduction is not necessarily optimal for energy consumption, and vice versa. Low Carbon Virtual Private Clouds 3

Outline  Introduction  Previous Work  Clean Energy Efficiency Model in a VPC: LCVPC Model  Simple Use Case for the LCVPC Model  Conclusion and Future Work Low Carbon Virtual Private Clouds 4

Introduction  Cloud computing solutions enable small businesses to rent virtual servers as a service, instead of buying and maintaining actual servers.  Apart from cost of services, due to increasing concerns about global warming and the increasing role of Greenhouse Gases (GhG) emissions, the total carbon footprint of a service is also of great concern to companies and governments. Low Carbon Virtual Private Clouds 5

 Hence, it is important for cloud service providers to be able to provide their customers with measurable proof of the carbon footprint of their services. Low Carbon Virtual Private Clouds 6

Virtual Private Cloud (VPC)  A Virtual Private Cloud is a uniform cloud based on a number of geographically distributed data centers which are connected through the Internet or private WAN connections. Low Carbon Virtual Private Clouds 7

Virtual Private Cloud  Network of Data Centers in different domains.  Connected to one another through private WAN connections or via the internet.  Each Data Center is powered by a different energy source.  Each Data Center is situated at a different geographical location. Low Carbon Virtual Private Clouds 8

LAN Based Clouds  Completely isolated data center.  Limited by their geographical location.  Powered by the same energy source. Low Carbon Virtual Private Clouds 9

Virtual Machine Migration for energy efficiency  LAN Based live VM migration – Cloud Administrators were able to move a VM from one hardware setup to another in the same Data Center, for maintenance or energy efficiency reasons without violating the Service Level Agreement.  WAN Based live VM Migration – Moving a VM from one Data Center to another Data Center. Recent research as proven that VM Migration over a WAN is also feasible for a Virtual Private Cloud.  The main idea of VM Migration is to consolidate VMs as much as possible. Low Carbon Virtual Private Clouds 10

Low Carbon Virtual Private Clouds 11 Virtual Machine Consolidation Reduce the amount of in-use hardware Save Energy Reduce the Carbon Footprint Virtual Machine Consolidation

Outline  Introduction  Previous Work  Clean Energy Efficiency Model in a VPC: LCVPC Model  Simple Use Case for the LCVPC Model  Conclusion and Future Work Low Carbon Virtual Private Clouds 12

Previous Work  Kansal’s model for VM Power Metering E sys = α cpu µ cpu + α mem µ mem + α io µ disk + γ(1) E sys – Energy Consumption of a Server. α – Additional Energy consumption of the server under 100% CPU, Memory or Disk usage. µ – Actual Percentage of CPU, Memory or Disk usage. γ – Energy Consumption of the server under 0% CPU, Memory or Disk usage. Low Carbon Virtual Private Clouds 13

 Gmach’s work on Dynamic VM Consolidation  Move VMs as much as possible from low-use servers and then turn those servers off to save energy.  Cost Function of VM Migration: cost = C(Migration) + C(#PM) + C(Utilization) (2) C(Migration) – Cost of VM Migration. C(#PM) – Cost of Physical Machine energy consumption. C(Utilization) – Cost of server use, which shows how busy the servers are. Low Carbon Virtual Private Clouds 14

 Merwe’s WAN VM Migration Design  Follow-the-Sun Scenario : Resources are relocated seamlessly to the place where they are needed most, based on the time zone.  In this paper, a similar design is used but migration is based only on the cost function which is set to reduce the network’s carbon footprint. Low Carbon Virtual Private Clouds 15

Outline  Introduction  Previous Work  Clean Energy Efficiency Model in a VPC: LCVPC Model  Simple Use Case for the LCVPC Model  Conclusion and Future Work Low Carbon Virtual Private Clouds 16

Clean Energy Efficiency Model in a VPC: LCVPC Model  A VPC Manager is used to optimize the location of VMs in the cloud based on the availability of resources and the carbon footprint of each data center.  Data Centers are situated at different locations and are powered by different energy sources.  A Data Center can be powered by:  Renewable Energy Source – Smaller or Zero Carbon Footprint.  Non-Clean Energy Source – Bigger Carbon Footprint.  Move some VMs from non-clean powered Data Centers to cleaner or totally clean powered Data Centers if they are available. Low Carbon Virtual Private Clouds 17

Power Consumption and Carbon Footprint for a Data Center  Carbon Footprint is directly proportional to Power Consumption  For a Data Center powered by a completely Clean Energy Source ` C p d (t) = 0  For Data Centers powered by any other Energy Source C p d (t) – Amount of Carbon emitted from data center d in time t. – Power-to-carbon conversion rate of data center d in time t. – Power Consumption of data center d in time t. Low Carbon Virtual Private Clouds 18

Cleanness Factor  Every Data Center could be powered by different energy sources and every energy source has its own carbon footprint.  The Cleanness of all energy sources can be represented by, g d (t) ε [0,1] where 1 represents totally clean energy and 0 represents a totally non-clean energy source.  Thus, Power-to-carbon conversion rate can also be expressed as ρ max represents the carbon-to-power conversion rate for the least clean energy source. Low Carbon Virtual Private Clouds 19 Power-to-carbon conversion rate when there are multiple energy sources at the same data center

Carbon Foot Print of the Cloud Low Carbon Virtual Private Clouds 20 Sum of the Carbon Footprints of all Data Centers Power Consumed by a data center comprises of power consumed by cooling, power processed by the Power Distribution Unit, and the power consumed by the servers. From Equation (1)

Low Carbon Virtual Private Clouds 21 Combining equations (6) and (8) provides the carbon cost function for a Virtual Private Cloud A Final Cost function by combining all the Carbon Footprint Formulas

 O d and O s – Binary variable and is equal to 1 when the data center or server is functional and is equal to 0 when the data center of server is shut down.  Δ t – Period of time where power measurements are constant or with small variations.  If there is no VM running at a data center or on a server, the data center or server could be shut down in order to eliminate the power consumption for cooling and overhead.  However, there is a Carbon Footprint generated to shut down a data center and to turn it back on which is considered in the C DC on/off Low Carbon Virtual Private Clouds 22

Outline  Introduction  Previous Work  Clean Energy Efficiency Model in a VPC: LCVPC Model  Simple Use Case for the LCVPC Model  Conclusion and Future Work Low Carbon Virtual Private Clouds 23

Simple Use Case for the LCVPC Model Low Carbon Virtual Private Clouds 24 Consolidate VMs at the Cloud Level WAN Migration of VMs is allowed in this step. Calculate Carbon Footprint. Consolidate the VMs at the Data Center Level Reduce carbon footprint. Calculate Carbon Footprint. A Set of VMs assigned to every data Center randomly Calculate Carbon Footprint for the whole network for a period of 24 hours. Compare LAN-Based Server Consolidation with WAN-Based Server Consolidation – Test Scenario 1.

Simulation Platform  7 Cities around the world  13 Data Centers  Different Energy Source for each Data Center.  Real Simulation Data  Geographical Coordinates  Sun’s Positions  Randomly Generated Simulation Data  Wind Stream Movements Low Carbon Virtual Private Clouds 25

At each Data Center the source(s) of energy and the power use percentages are provided separately. Low Carbon Virtual Private Clouds 26

Parameters of a Typical Data Center in a VPC Low Carbon Virtual Private Clouds 27

Network Carbon Footprint  Variety of energy sources.  Simulation for 72 hours.  Maximum Carbon Footprint is reached at hour 13 when the sun is in the middle of the Pacific Ocean from where it can no longer power any of the solar sites.  Also, there is no major wind stream near any of the wind sites. Low Carbon Virtual Private Clouds 28

Simulation Platform Map at Hour 13 Low Carbon Virtual Private Clouds 29

Optimization  A Genetic Algorithm (GA) is used to Optimize the network.  The GA optimizes of Energy as well as for Carbon Footprint Low Carbon Virtual Private Clouds 30 Genetic Algorithm Decide which servers to consolidate and which servers to turn off. Removal of all VMs from a non-clean server and their migration to other, preferably green severs. Optimize Carbon Footprint while considering available memory, CPU, and storage on each server. Decide which Data Centers needs to be turned off and migrate all its VMs to other Data Centers.

Network Carbon Optimization under large Intervals Low Carbon Virtual Private Clouds 31

Network Carbon Optimization under small Intervals Low Carbon Virtual Private Clouds 32 The optimum interval value lies between 0.5 and 2 hours. 40 servers were run 13 data centers. 7 different cities.

Simulation under different VM Loads – Test Scenario 2  Heavy Load – 60% CPU usage.  Light Load – 33% CPU Usage.  Normal Load – 47% CPU Usage. Low Carbon Virtual Private Clouds 33

Network Carbon Footprint under heavy VM Load  60% CPU Usage Low Carbon Virtual Private Clouds 34

Network Carbon Footprint under light VM Load  33% CPU Usage Low Carbon Virtual Private Clouds 35

The Energy Equation – Test Scenario 3 Low Carbon Virtual Private Clouds 36

Network Energy Measurement Low Carbon Virtual Private Clouds 37

Network Carbon Measurement Low Carbon Virtual Private Clouds 38

Outline  Introduction  Previous Work  Clean Energy Efficiency Model in a VPC: LCVPC Model  Simple Use Case for the LCVPC Model  Conclusion and Future Work Low Carbon Virtual Private Clouds 39

Conclusion and Future Work  The results show that VPC Data Consolidation has a more significant Carbon Footprint Reduction than LAN Server Consolidation.  The authors also conclude that carbon reduction is not necessarily equal to energy efficiency in VPCs.  For future work, use cases can be testing by varying the VM Load, increasing the number of cities Low Carbon Virtual Private Clouds 40

 Instead of using randomly generated wind stream data, real wind stream data could be used to generate more realistic results.  For a more realistic solar energy simulation, simulated clouds can be considered in the simulation platform.  In conclusion, greater carbon footprint reduction can be achieved through reduced power use in a VPC.  However, this may not be profitable for investors in VPCs because of the correspondingly reduced use of their infrastructure. Low Carbon Virtual Private Clouds 41

Thank You! Low Carbon Virtual Private Clouds 42