Incorporating Blockchain into RDF Store at The Lightweight ... … · –reduce network overhead. –reduce latency for real-time applications. –improve scalability. –better privacy

Incorporating Blockchain into RDF Store at The Lightweight Edge Devices

Anh Le-Tuan, Darshan Hingu, Manfred Hauswirth and Danh Le Phuoc | Technische Universitat Berlin

RDF4LED - An RDF Engine for Edge Devices

Incorporating Blockchain into RDF4Led

Incorporating Blockchain into RDF Store at The Lightweight Edge Devices | Anh Le-Tuan |

Seite 2

Why Edge Devices ???

Edge Computing - Internet of Things

– reduce network overhead.

– reduce latency for real-time applications.

– improve scalability.

– better privacy control.

Linked Data

– enable data integration of heterogeneous sources.

– enable data federation over edge nodes.


Seite 3

Missions

Moving semantic data processing task away from centralised cloud for the IoT

– How much semantic data on small devices ?

– How to scale data federation over small devices on edge systems ?

– How to encourage people sharing the data from their edge devices ?


Seite 4

RDF4LED - An RDF Engine for Edge Devices

System Design


Seite 5

RDF4LED (cont.)

Physical RDF Organisation: Two-Layers Index


Seite 6

RDF4LED (cont.)

Targeted Small Devices


Seite 7


System Overview

Smart Contract Manager

Distributed RDF Storage


Seite 8


System Workflow

Smart Contract Manager

SPARQL Query Processor

Validation Service

SmartContracts

Storage

Trip

le P

atte

rn

Req

uest

Smart Contracts

2

Payment

Request P

aym

ent

Con

firm

atio

n

Trigger Contracts

Index Entries

Index Entries


Req

uest

with

In

dex

Entr

ies

Matched

Triples

3

45

1

6

7

8

9


Seite 9


Physical Organisation

. . .Transaction N: {Owner: ClientN Price: 0.1 IPFS:hashN}

Transaction 2: {Owner:ClientB Price: 0.1 IPFS:hash2}

. . .IPFS:hash1s1 p1 o1 o2 o3; p2 o4 o5p3 o6.

IPFS:hash2s2 p1 o5; .... .. .. .... .. .. ..

IPFS:hashNs3 p3 o2; .... .. .. .... .. .. ..

Transaction 1:{Owner: ClientA Price: 0.1 IPFS:hash1}

. . .s2p1o5:{ Transaction 2}

s3p3o2:{ Transaction N}

s1p1o1:{ Transaction 1}


Blockchain

Smart Contracts Storage


Seite 10


System Deployment

ValidationService

Client A

Distributed Storage System Cluster

Miner Nodes

Geth Clients

Client B Client N

....


Seite 11


Evaluation - Input (1) Acc. Throughput on Static Cluster Sizes

0 200 400 600 800 1,0001,000

1,500

2,000

2,500

DatasetSize (in millions)

Thro

ughp

ut(tr

iple

s/se

c)10 nodes Inserting


Seite 12


Evaluation - Input (2) Acc. Throughput on Varying Cluster Sizes

2 4 6 8 10

10

20

30

40

No of Nodes

Thro

ughp

ut(tr

iple

s/se

c)in

thou

sand

s100 Million triples Dataset


Seite 13


Evaluation - Query

0.5

5

3060

350

1000

1 10 50 100 500 1000

seco

nd

(in

log

sca

le)

Number of blocks return in query patterns


Seite 14

Thank You !!!


Seite 15

RDF4LED (cont.)

Write Management (1): Clustering example


Seite 16

RDF4LED (cont.)

Write Management (2): Cache Management


Seite 17

RDF4LED (cont.)

Push-based Join Algorithm


Seite 18

RDF4LED (cont.)

Evaluation Results - Input (1)


Seite 19

RDF4LED (cont.)



Seite 20

RDF4LED (cont.)



Seite 21

RDF4LED (cont.)

Evaluation Results - Query (1)

1

10

100

F1 F2 F3 F4 F5 L1 L2 L3 L4 L5 S1 S2 S3 S4 S5

seco

nd

(in

log s

cale

)

Queries

(a) Query response time on Intel Galileo Gen II

VIRTUOSORDF4LedJENA TDB


Seite 22

RDF4LED (cont.)


1

10

100


seco

nd

(in

log s

cale

)

Queries

(b) Query response time on Raspberry Pi Zero



Seite 23

RDF4LED (cont.)


1

10

100


seco

nd

(in

log s

cale

)

Queries

(c) Query response time on Beagle Bone Black



Seite 24

RDF4LED (cont.)


1

10


seco

nd

(in

log s

cale

)

Queries

(d) Query response time on Beagle Bone Black (20 mil)

VIRTUOSORDF4Led


Seite 25

RDF4LED (cont.)

Evaluation Results - Memory (1)


Seite 26

RDF4LED (cont.)

Evaluation Results - Memory (2)


Seite 27

Documents

Incorporating Blockchain into RDF Store at The Lightweight ... … · –reduce network overhead. –reduce latency for real-time applications. –improve scalability. –better privacy