Storage device performance prediction with CART models(19)

Storage device performance prediction is a key element of self-managed storage systems and application planning tasks, such as data assignment. This work explores the application of a machine learning tool, CART models, to storage device modeling. Our appr

12000

Number of Requests

9000

6000

3000

Time (aggregated in 10 seconds)

Disk Blocks (aggregated in 1000 blocks)

20

Disk block number

Entropy value

15 10 5 0

Entropy on timeEntropy on LBNJoint entropyCorrelation

60000

40000

20000

Arrival time

0 5

10Scale

15 20

Entropyplotonone-dimensionaldatasets.Theone-dimensionalentropyplotcapturescharacteristicsofindividualattributes,suchasthetemporalandspatialburstiness.Thesetwotypesofburstinessmeasurestheburstinessinthearrivalprocessandtheskewinaccessfrequenciesofdiskblocks.Weusetheentropyplotforarrivaltimeasanexampletoshowhowtheentropyplotworks.

Givenaworkload,wecanderiveits“margin”onthearrivaltimebycountingthenumberofrequeststhatarriveintothesystemateachtimetick.ThetopgraphofFigure9(a)showsthesampletrace’smarginonarrivaltime.

2n.WecalculatetheAssumethatthetraceis2ntimetickslong,andthemarginisCii12

entropyvalueatscalekbyapplyingtheentropyfunctionontheaggregatedmarginatscalek.TheaggregatedmarginisCkjj122kwhere

Intuitively,theentirelengthofthemarginisdividedinto2kequi-lengthedintervalsatscalek.Thus,applyingtheentropyfunctiononCkgives

where

Pkj

Number of Requests

(a)Sampledisktrace(b)Entropyplot

Figure9:Asampledisktraceanditsentropyplot.

C

k

2n

k

j

i1

∑C2n

k

j

1

i

H

k

2n

k

∑Pkjlog2Pkj

j1

C

k

j