Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Capacity Planning Methodology(A Refresher)
Copyright: Pedro Escosteguy Resources Inc.CMG CANADA APRIL 2017
Objective To present a Capacity Planning
Methodology that has successfully been used in the following:
– UNIX environments: AIX, LINUX, HP UNIX, HANA, EXADATA, TERADATA
– Companies: Toyota, Eaton’s, BNS, CGI, Rogers, Loblaw
Copyright: Pedro Escosteguy Resources Inc.
Objective
Agenda
Introduction What, why and how to measure The approach The process Forecasting Summary
Copyright: Pedro Escosteguy Resources Inc.
Introduction
Problems managers face: Thousands of machines Tools costing hundreds of dollars per license
per machine resulting in millions of dollars in cost
Need a process that is simple, fast, easy to understand, and has a Low Cost
Have a budget to manage, so don’t have lots of money to spend with expensive tools
Copyright: Pedro Escosteguy Resources Inc.
What to Measure
Workload characterization
Total CPU consumption by measuring CPU utilization time (not percentage)
Copyright: Pedro Escosteguy Resources Inc.
Why Measure CPU Utilization
CPU is a CPU is a CPU
If everybody waits at the same speed, everybody uses CPU at the same speed. The difference is for how long.
Copyright: Pedro Escosteguy Resources Inc.
Sealed Box Model
A way of measuring CPU performance
How the Box works is of no interest
It measures the work flowing through the Box
[TM] Introduction to Large Processor Capacity and Performance Evaluation -IBM manual GG 66-0232, May 1986 by Mr. Kenneth L. Radecki
Sealed Box Model (cont.)
ITR (Internal Throughput Rate), ETR (External Throughput Rate), IER (Internal Execution Rate)
IBM Data Processing Techniques manual, Analysis of Some Queuing Models in Real-Time Systems [GF 20 - 0007]
Queuing Theory Equations
Response Time
waiting processing
TQ
TW TS
RtLW
LQ
TQ = Tw + Ts 1.1
Queuing Theory Equations (cont.)
Queuing Theory Equations (cont.)Server Utilization Ut = Rt x Ts 1.2
Mean Queue Size Lq = Ut 1.31 - Ut
Mean Queuing Time Tq = 1 x Ts 1.41 - Ut
Mean Waiting Time Tw = Ut x Ts 1.51 - Ut
orTw = Lq x Ts 1.6
WhereRt = transaction rateTs = time to serve the transaction
figure 4
Queuing Theory Equations (cont.)
rule-of-thumb of the rules-of-thumb
Utilization = K 1.7K + 1
figure 5
Where K is average queue size
Queuing Theory Equations (cont.)ROTs:
Single path channels should be used less than 35%;
Double path channels can go up to 50%;
Four path channels can go up to 70%;
Eight path channels can go up to 80%;
Queuing Theory Equations (cont.) Single path channels should be used less than 35%;
channel 0 1 min 0, max 1, average 1/2 ½ / (½ + 1) = 1/3 = 33%
Double path channels can go up to 50%;channel 0 1 min 0, max 2, average 1channel 0 1 1/(1+1) = ½ = 50%
Four path channels can go up to 70%;channel 0 1 min 0, max 4, average 2channel 0 1 min 2/(2+1) = 2/3 = 67%channel 0 1channel 0 1
Eight path channels can go up to 80%; 4/5 = 80%
Disk
disk
disk
CPU 1
00.10.20.30.40.50.60.70.80.9
1
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58
CPU 2
00.10.20.30.40.50.60.70.80.9
1
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58
CPU 3
00.10.20.30.40.50.60.70.80.9
1
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58
CPU 4
00.10.20.30.40.50.60.70.80.9
1
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58
CPU 5
00.10.20.30.40.50.60.70.80.9
1
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58
CPU 6
00.10.20.30.40.50.60.70.80.9
1
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58
Which chart usedTheMost
CPU?
CPU 6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 1 2 3 4 5 6 7 8 9 1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859
How to Measure
vmstat
iostat
sar
ps (*)
Copyright: Pedro Escosteguy Resources Inc.
Additionally in AIX
SAR in LINUX does measure memory utilization. In AIX it does not
Topasrec and nmon do measure memory used and free
Copyright: Pedro Escosteguy Resources Inc.
The Approach
Measure resource (CPU) Tabulate the data Apply growth rate Produce chart Present to the manager
Copyright: Pedro Escosteguy Resources Inc.
The Process Crontab to schedule data collection
Scripts to reduce the data
Normalize the data
Produce forecast
Incorporate seasonality Copyright: Pedro Escosteguy Resources Inc.
DEMO
daily profile (1)Server A
Copyright: Pedro Escosteguy Resources Inc.
daily profile (2)Server A
Copyright: Pedro Escosteguy Resources Inc.
monthly utilization
Server A
Copyright: Pedro Escosteguy Resources Inc.
Forecasting
Data regression
Businesses growth
Copyright: Pedro Escosteguy Resources Inc.
Forecasting (cont.)
X variable 1 line Fit Plot
0
5
10
15
20
25
30
Jan-
94
Feb-
94
Mar
-94
Apr-9
4
May
-94
Jun-
94
Jul-9
4
Aug-
94
Sep-
94
Oct
-94
Nov
-94
Dec
-94
Jan-
95
Feb-
95
X variable 1
Y
Predicted Y Y
Copyright: Pedro Escosteguy Resources Inc.
data regressionServer A
y = b + m * x Copyright: Pedro Escosteguy Resources Inc.
seasonalityServer A
Copyright: Pedro Escosteguy Resources Inc.
Capacity Planning
Methodology(A Refresher)
Copyright: Pedro Escosteguy Resources Inc. (416) 258-5341
CMG CANADA APRIL 2017
SUMMARY