Andrew Wang (andrew.wang@cloudera.com, @umbrant) Colin McCabe (cmccabe@cloudera.com)

This repository contains benchmarks, scripts, and data related to our Hadoop Summit Amsterdam 2014 presentation, "In-memory Caching in HDFS: Lower latency, Same great taste."

• micro: vecsum2 microbenchmark, libhdfs program used to test raw throughput of different HDFS read methods
• mapreduce: ByteCount MapReduce job which performs zero-copy reads
• graphs: flamegraphs and plotting scripts from vecsum microbenchmark
• scripts: collection of helper scripts used to run multi-node experiments with MapReduce and Impala, including the SQL queries used. Also contains raw data and parsing scripts used to make our graphs.

Everything here is under the Apache License, Version 2.0. We're in the process of contributing some of the benchmark code back to core Hadoop.