Paper Title: Analyzing & Characterizing Enterprise Workloads at Block and Driver layers

Paper Abstract: In this session we will discuss an end-to-end process of analyzing and characterizing the I/O of enterprise workloads. For data collection we use a mix of well-known methods and lesser-known methods (blktrace, bpf & perf) to collect Linux block layer events and Linux NVMe driver events. These events are processed with simple scripts to produce dataframes that can be analyzed in any standard data analysis framework (pandas, spark, R, etc). These processes enables high level analysis to characterize workloads as well as collection & analysis of NVMe specific data without the need for a bus analyzer. By using standard Linux tools, this process enabled users and developers to gain deeper insights into their workloads. Finally, we will discuss sample analyses of block & nvme event data to understand workloads and system performance.

Paper Author: Wesley Vaske, Principal Storage Solutions Enginee, Micron Technology
Sundararajan Sankaranarayanan, SSD Systems Architect, Micron Technology

Author Bio: Wes Vaske is a Principal Systems Performance Engineer on the Micron Data Center Workloads Engineering team in Austin Texas. He and his team analyze enterprise workloads to understand the performance effects of Flash and DRAM devices on applications as well as to provide 'real-life' workload characterization to internal design & development teams. Wes's specific focus on is deep workload analysis and developing the tools to extract new observations. Prior to joining Micron in 2015, Wes was a Systems Engineer at Dell on the Oracle Solutions team focused on performance analysis of Oracle Database workloads.

Author 2 Bio: