Presentation Gave at IBMSTU 2017 ( São Paulo )

Purpose

This is a simple DEMO idea of how to use Linear Regression to estimate server grownth over time It use nmon data to predict size over time At current stage it predict:

• Estimated date to reach 100% of entitled capacity ( IBM POWER LPAR )
• Estimated date to reach 95% of CPU usage
• Amount of cores being used when 95% CPU is reached

Requirements

• Pymongo ( >= 3.2 )
• TensorFlow ( >= 1.0 )
• Python ( >= 3.5 ) ( Should work with 2.7, but wasn't tested )

Usage

The idea is composed by two scripts

nmon_parser.py :

This script is responsible to parse all nmon files presented at one directory. It will scan the directory for nmon files ( for a specific LPAR ) and parse them in parallel The following fields are being considered during Parsing:

• AAA | Basic information
• BBBL | Lpar configuration
• ^PCPU_ALL$|^SCPU_ALL$|^CPU_ALL$|^LPAR$|^MEM$|^PROC$|^NET$|^NETPACKET$|^NETSIZE$|^NETERROR$|^IOADAPT -> General Counters

The information about database connection and directory to be scanned are at the bottom of the script, bellow main reference For STU the following parameters were used:

directory = "/home/pqueiroz/clientes/IBM/STU_2017/dset/nmon"
mongodb_uri = 'mongodb://localhost:27017'
mongodb_db = 'stu2017'
paralel = 1

• directory : Control which directory the nmon files are stored
• mongodb_uri : uri to connect at the mongo database ( standard from mongodb )
• mongodb_db : Database name to be used
• paralel : Amount of connections in Parallel

predict_size.py

With all Data inserted at rhe database, this script will predict the sizes and dates All parametrization is at top of the script and those are the parameters used at STU:

learn_rate = 0.001
samples = 100000
mongodb_uri = 'mongodb://localhost:27017'
mongodb_db  = "stu2017"

TODO

• Use argparse or getopt to avoid change the code
• Use pyplot for visualization
• split the dataset to consider testing