Description

Abstract—Efficiently  analyzing big data  is a major issue in our current  era.

 Examples of analysis tasks include  identification  or detection of global weather

patterns, economic changes, social phenomena, or epidemics.

The cloud computing  paradigm along  with software tools such  as

implementations of the popular  MapReduce framework  offer a response to the 

problem  by distributing  computations among  large  sets of nodes. In many

 scenarios input data  is however geographically distributed  (geo-distributed )

across datacenters, and  straightforwardly moving  all data  to a  single  datacenter

before  processing it can  be  prohibitively  expensive.  Above-mentioned tools  are

 designed to work within a  single  cluster  or datacenter and perform poorly or

not at all when deployed across datacenters.

This  paper deals with executing sequences  of MapReduce jobs  on  geo-distributed datasets.

 We  analyze possible ways  of executing such  jobs,  and  propose data  transformation graphs

 that  can  be  used to determine schedules for job sequences which are  optimized  either  with

respect to execution time or monetary cost.  We introduce G-MR, a system for executing such

job sequences, which implements our optimization  framework.