Saturday, 28 November 2015

Some Notes from Running a Perfstat against a busy cDOT Cluster

Note i: This post is for the CLI version but there is also a GUI version.
Note ii: As always, this is just a blog post and not support material - the switches may not be optimal for your situation.

1) Make sure you download the right and latest version of PerfStat (which is currently 8.3):


2) Running perstat8.exe:

For scanning one node:

perfstat8.exe --mode=c --nodes={NODE_IP_ADDRESS} -z -t 3 -i 10,0 --preset-file=preset_minimal.txt --exclude="SHELL=SYSTEMSHELL"

Note iii: This is 10 iterations of 3 minutes with 0 gap between iterations, but against a busy node may take 1 hour +

Similarly, to perfstat the whole cluster:

perfstat8.exe {CLUSTER_IP_ADDRESS} --mode=c -z -t 3 -i 10,0 --preset-file=preset_minimal.txt --exclude="SHELL=SYSTEMSHELL"

Explanation of the switches involved:

--mode=c: cDOT Mode
-z: Only filer scan (i.e. don't scan hosts)
-t: Time in minutes (3 is optimal)
-i A,B: A iterations, with B minutes in between iterations (increase iterations to record a longer interval)
--preset-file: Used to exclude certain commands (so the Perfstat doesn't take forever)
--exclude="SHELL=SYSTEMSHELL": skips needing the diag user login

Image: Perfstat8 in action
Contents of the Preset_Minimal.txt File

######################
# Perfstat Preset File
# SECTION       TYPE    HARDWARE        OS      LICENSE SHELL FLAG   SHORTNAME   COMMAND
PRESTATS      CONFIG NODE   >=8.1  *      NGSH                 volume efficiency policy show
PRESTATS      PERF   NODE   >=8.1  *      NGSH                 volume efficiency stat
PRESTATS      PERF   NODE   *      *      NODESHELL                  wafl_susp -z
PRESTATS      PERF   NODE   *      *      NGSH                 statistics show -node local -category latency -object latency -counter nfs-histogram
PRESTATS      PERF   NODE   *      *      NODESHELL                  waffinity_stats zero
PRESTATS      PERF   NODE   *      *      NODERUN              stats start   stats start
PARALLEL      PERF   NODE   *      *      NODERUN              sysstat_M     sysstat -M -c
PARALLEL      PERF   NODE   <8 -c="" -x="" .1.1="" nbsp="" noderun="" sysstat="" sysstat_x_1sec="" time=""> -s 1
PARALLEL      PERF   NODE   >=8.1.1       *      NODERUN              sysstat_x_1sec       sysstat -x -d -c
PARALLEL      PERF   NODE   *      *      NODERUN              statit stutter statit
PARALLEL      PERF   NODE   >=8.0 and <8 -category="" -interval="" -iter="" -node="" .1x="" 1="" local="" nbsp="" nfs3="" ngsh="" periodic="" statistics="" time="">
PARALLEL      PERF   NODE   >=8.1 and <8 -interval="" -iter="" -node="" -object="" .1.1="" 1="" local="" nbsp="" nfs3="" ngsh="" periodic="" statistics="" time="">
POSTSTATS     PERF   NODE   *      *      NGSH                 statistics show -node local -category latency -object latency -counter nfs-histogram
POSTSTATS     PERF   NODE   *      *      NODESHELL                   wafl_susp -w
POSTSTATS     PERF   NODE   *      *      NODESHELL                  wafl_susp -r
POSTSTATS     PERF   NODE   *      *      NODESHELL                  wafl scan status
POSTSTATS     PERF   NODE   *      *      NODESHELL                  wafl scan status -A
POSTSTATS     PERF   NODE   >=8.1  *      NGSH                 volume efficiency stat
POSTSTATS     PERF   NODE   *      *      NODERUN              stats stop    stats stop
POSTSTATS     PERF   NODE   *      *      NODESHELL                  waffinity_stats

Bonus Material - Instructions to FTP out via a Proxy Server

From DOS prompt on the server:

ftp YOUR_PROXY_SERVER

User: anonymous@ftp.netapp.com
Password: YOUR_EMAIL_ADDRESS

ftp> bin
ftp> hash
ftp> cd to-ntap
ftp> put {casenumber}.{filename}

Note iv: In Windows you can drag the file into the command line window from Explorer to paste the filename (the file needs to be on a local drive)
Note v: Instead of using ftp.netapp.com you can use ftp-europe.netapp.com

No comments:

Post a Comment