An Experimental Survey on Big Data Frameworks

1 . An Experimental Survey on Big Data Frameworks W. Inoubli, S. Aridhi, H. Mezni, M. Maddouri, E. Mephu Nguifo XLDB 2017 October 10-12, 2017 Casino de Royat, Allée du Pariou, 63130 Royat, FRANCE 10/10/2017 

2 . Context and motivations Big Data problems:  Scalability and fault tolerance requirements  Several applications have been migrated to Big Data  Several Big Data frameworks have been proposed XLDB October 10-12, 2017 Clermont- Ferrand, France 10/10/2017 2 

3 . Experimental protocol Batch Mode evaluation StreamMode evaluation  Workload: Kmeans, WordCount and PageRank.  Frameworks: Hadoop (Mapreduce), Spark and  Workload: ETL Workload. Flink.  Frameworks: Storm, Spark and Flink.  Features: Scalabilty, Configuration parameters.  Features: Number of events processed. XLDB October 10-12, 2017 Clermont- Ferrand, France 10/10/2017 3 

4 .Experimental protocol Monitoring Tool Data collection: kafka. Data Strorage: Elastic search. Data Visualisation: Kibana 10/10/2017 XLDB October 10-12, 2017 Clermont- 4 Ferrand, France 

5 . Experimental results Batch Mode results : (a) Small datasets (b) Big datasets Impact of size of data on response time, with small and big datasets. 10/10/2017 XLDB October 10-12, 2017 Clermont- 5 Ferrand, France 

6 . Experimental results Batch Mode results :  Vary the number of machines in cluster.  kmeans workload  Study the scalability feature.  Vary the number of iterations, 10/10/2017 6 XLDB October 10-12, 2017 Clermont-Ferrand, France 

7 . Experimental results Stream Mode results : Event with 100 kb Event with 500 kb Impact of the size of messages on the number of processed messages  Storm performs better in the case of small datasets.  Flink provides good results in the case of big datasets. 10/10/2017 XLDB October 10-12, 2017 Clermont-Ferrand, 7 France 

8 .More at the Poster Session ! 10/10/2017 XLDB October 10-12, 2017 Clermont-Ferrand, 8 France