Application-Network Tracing and Correlation in Datacenters (ANTACID) Plan of Action • Collect packet traces + application logs – MapReduce as case study – Run at Yahoo! If possible, else on EC2 • Correlate low level and high level logs – What do app logs tell us about network? – How network affects app performance? • Leverage Chukwa data collection tool – Collection of distributed packet-traces – Centralized + scalable storage on HDFS – Simplifies analysis and visualization A. Rabkin & A. Konwinski, UC Berkeley Impact • Simple tools for collecting and analyzing large network level traces will be rapidly adopted by many corporations, such as Yahoo!. • Collection, organization, and storage of low level trace data will facilitate work by other researchers. • Availability of data and analysis tools (through Chukwa) will facilitate new uses of the data, e.g. replayable traces. Schedule • Current: Matrix visualization of data collected on 7 node Hadoop cluster. • Oct 10: Collect packet traces on 7 node cluster • Oct 23: Discuss Chukwa integration at CCA • Nov 4: Collect large traces and logs • Dec 9: Project poster