Bab 5: Workload Characterization Techniques - Bab 5 Workload Characterization - Repository UNIKOM

Bab 5: Workload Characterization Techniques

  1 Workload Characterisation

  ! " # #

  

Techniques for Workload Characterization

  $ % & ' ( ' )

• , ,

  ! -

  3 AVERAGING

  % "

SPECIFYING DISPERSION

  & .

  " ! ! % ! / .

  5 Case Study 6.1 The resource demands of various programs executed on six university sites were

measured for 6 months. The average demand by each program is shown in Table 6.1. Notice that

the C.O.V. of the measured values are rather high, indicating that combining all programs into one class is not a good idea. Programs should be divided into several classes. Table 6.2 shows

SINGLE-PARAMETER HISTOGRAMS

  % % # # #

  1

  2

  3

# " "

"

  4

  14

2 5/ !+6

4 %

  4 !+6 5/ 7 ₇ MULTIPARAMETER HISTOGRAMS

  %

  1

  2

PRINCIPAL-COMPONENT ANALYSIS

  : $ "

  " #

  $ & (

  8

  8 ( "

  8 " #

  & "

  8 ; "

  8

  / #

  6 " /

  9

  :

  $ &

  9

  '

  9

  #

  8 "

  8

  $ &

  #$ % & " "

  " , ) % , & %

  "

  11

• (*& _? ^? _#

  '

  A *

  &

  '

  &

  ( )

  &

  A $ >$, '

  $

  @

  &

  A = " $ >$, = =<) , % "

  &

  = >$,

  &

  # # 6 ) 6 ) 1@ $2

  $ % & ' ( % "

  $(<= )<<> $-)$

  " , ) ¹³

  8 "

  8

  "

  ' ! " "

• $$

  &$ B $

  $ & _$

• $ &

  '

  $ & _&

• $ &

  , % + + , % % % " " _{$ . . ?} $ $ *(*& $-)$

  & & _{$ &} * * _{? &} $ $ )<<> & & $(<=

  " - " , )

  15

• /

  " " #

  # (& *,* $ )(* " (& *,*51(& *,* C $ )(*2 >* -D

  " ) (D 0 '

MARKOV MODELS

  17 #$ % * " 7 " <=D

  % <=D " & %

  % # # %

  "

• %

  = $ # = < E " "

  

CLUSTERING

  " $ " & ' ( " ) B

• ' , '
• < > !

  ( - $= ' ₁₉

  & + " /

  " ,

  " . + "

  ! "

  1

  2

  3 4 % 5 6 $ 7 % ! " & ' ! "

  

Distance Metric

  8

  9 4 %

  8

  9 $ : *

  9

  21 Clustering Techniques

  "