Benchmarking,Analysis And Optimization Of In-Memory Databases

Energy consumption has become a fundamental design constraint for high end distributed systems.Energy cost plays an important role for successful operation of any data center.Conventional disk based Database systems which is the underlying driver behind online services undoubtedly plays significant contribution in consuming noticeable amount of energy when executing a single query partly due to their high access latency.In-memory database systems(IMDSs)that keep data all the time in Random Access Memory(RAM)hold out the promise of breakthrough energy performance for time-sensitive,data-intensive tasks by improving system function greatly due to its low access latency.There are various in-memory databases currently in existence,such as:Aerospike,Store,RavenDB,Altibase,HyPer,Redis,Clusterpoint,HyperSQL,SAPHANA,CSQ,kdb+,solidDB,DataBlitz,MDB,SQLite,eXtremeDB,MemSQL,Starcounter,FastDB,MonetDB,Ti-mes Ten,Hekaton,Polyhedra,VoltDB etc.Due to emergence of such many companies into in-memory database technology,there is need for an independent benchmark to evaluate both performance and energy consumption of their databases under different business scenario.This thesis therefore brings forth a revolutionary unified simple to use,powerful,flexible and intuitive API for a comprehensive query based energy-performance benchmark for both distributed homogenous and heterogeneous in-memory databases.We shall profile the energy consumption for core server components(i.e.CPU usage,memory usage,network I/O etc.),and by varying workloads such as increasing or reducing the number of nodes in a cluster,increasing or reducing data amounts etc.,we shall assess the energy consumption of three in-memory databases(i.e.MySQL Cluster,Radis and MongoDB),in relation to their performance(measured by query time).