Evaluation of Information Service Architectures for Grid.

Post on 19-Jan-2016

213 views 1 download

Tags:

transcript

Evaluation of Information Service Architectures for Grid

MDS-4 and DHT

• Hyerarchical architecure

• Flat architecture

• Created for different uses

Focus

• Information systems for grid• Metadata: resources + users• Churn• Pragmatical approach

Constraints

• Fixed schema

• Multi-attribute query

• Range query

• 3 attributes

Index list search

• The list is long• The items are geographically dispersed

Hierarchy, MDS4

• Bottleneck for performance• Collected,cached,leased data.• + slow data flow, - old data

Peer-to-peer, Multi Chord

• + good for large number of keys

• - Churn

Experiment

• Size (#nodes, #items) EGEE 250 ,25000 • Schema [CPU,RAM,Diskst] [Group,Role,Capability]

• Selectivity on whole range

TTF,TTR skewed, ρ= 0.1, 250 nodes, 120 minutes,CPU data

,00

20,00

40,00

60,00

80,00

100,00

120,00

1/MTTF

Neg

ativ

e ra

te

MDS4

p2p

MDS4 43,87 37,84 49,61 62,48 59,67 89,48 70,49

p2p 23,78 45,90 75,00 63,00 73,00 73,00 96,30

0,016 0,002 0,025 0,033 0,050 0,100 0,500

Std. Dev.

0,00

20,00

40,00

samples

SD SD MDS4

SD p2p

SDMDS4

0,20 2,38 15,00 0,80 1,70 15,00 0,50

SD p2p 25,0019,0013,00 4,70 11,0032,00 1,29

1 2 3 4 5 6 7

TTF skewed,TTR fixed, ρ= 0.1, 250 nodes, 120 minutes,CPU data

,00

20,00

40,00

60,00

80,00

100,00

1/MTTF

Neg

ativ

e ra

te

MDS4

p2p

MDS4 41,00 47,00 50,00 58,00 62,00 68,00 70,00

p2p 46,00 32,00 58,00 63,00 65,00 92,00 68,36

0,016 0,002 0,025 0,033 0,050 0,100 0,500

TTF skewed,TTR fixed, Membership data

,00

10,00

20,00

30,00

40,00

50,00

60,00

70,00

80,00

90,00

1/MTTF

Neg

ativ

e ra

te

MDS4

p2p

MDS4 43,00 45,00 50,00 59,00 74,00 70,00 71,00

p2p 40 36 37 56 78 74 83

0,016 0,002 0,025 0,033 0,050 0,100 0,500

VOMS vs CPU dataChord: voms vs cpu

0

20

40

60

80

100

1/TTF

nega

tive

rate

p2p_voms

p2p_cpu 46,00 32,00

p2p_voms 40 36 37 56 78 74 83

p2p_cpu 46,0032,00

46,00 32,00 58,00 63,00 65,00 92,00 68,36

1/TTF

MDS4: voms vs cpu

,00

20,00

40,00

60,00

80,00

1/TTF

nega

tive

rate

gt4_voms 43,00 45,00

gt4_cpu 41,00 47,00

gt4_voms 43,0045,00

43,00 45,00 50,00 59,00 74,00 70,00 71,00

gt4_cpu 41,0047,00

41,00 47,00 50,00 58,00 62,00 68,00 70,00

1/TTF

Conclusions

• MDS4, ram, slow, churn<5min,

• Much better than Chord on skewed data

• Chord multi

• Fast

• Returns fewer keys

Future

• More info available on Dynamic Behaviour of nodes?

• Increase size? / Distributed simulation platform

• Data models

• Query composition: selectivity and bias

• How coul make analysis useful to cloud /hibryd solutions?

• Any other middleware to compare?

• Relevant for DHT ?

Thank you for being here!

Environment of the experiment

Schema - CPU processor data

Attribute Unit Type Distribution

Avg. Range:min - max

CPU MHz

integer Skewed 2000

RAM MB integer uniform 100-300

Disk Storage

GB integer uniform 400-1000

Schema – membership data

Attribute Unit Type Distribution

Avg.

Range:min - max

Group ID integer

uniform NA 1-20

Role ID integer

uniform NA 1-20

Capability

ID integer

uniform NA 1-20