To realise a semantic Web of Things, the challenge of achieving efficient Resource Description Format (RDF) storage and SPARQL query performance on Internet of Things (IoT) devices with limited resources has to be addressed. State-of-the-art SPARQL-to-SQL engines have been shown to outperform RDF stores on some benchmarks. In this paper, we describe an optimisation to the SPARQL-to-SQL approach, based on a study of time-series IoT data structures, that employs metadata abstraction and efficient translation by reusing existing SPARQL engines to produce Linked Data `just-in-time'. We evaluate our approach against RDF stores, state-of-the-art SPARQL-to-SQL engines and streaming SPARQL engines, in the context of IoT data and scenarios. We show that storage efficiency, with succinct row storage, and query performance can be improved from 2 times to 3 orders of magnitude. Presentation at International Semantic Web Conference 2016

1. INTERNET OF THINGS DEVICES ANDSTREAMS
SPARQL-TO-SQLON
EUGENE SIOW
THANASSIS TIROPANIS
WENDY HALL

2. Trekking up 913m
Mt. rokko

3. Sensor DATAFROMMULTIPLE SOURCES
“The Internet of Things is currently beset by product silos.”
W3C Web of Things Interest Group
CURRENT STATE OFTHE
INTERNET OF THINGS
PRODUCT &DATASILOS
DEPENDENCYON THE CLOUD
LIMITED INTEROPERABILITY OF APPS

4. LINKED DATA ON LIGHTWEIGHT COMPUTERS
APPLE
GOOGLE
HONEYWELL
CISCO
HUAWEI
GENERALELECTRIC
IBM
AMAZON
INTEL
MICROSOFT

5. INTRODUCING
LINKED DATA
FOR INTEROPERABILITY
URI andontologies
Establish commondata structures& References
ENABLES RICH METADATA
what,where, WHEN,HOW of DATA
Integrates with Linked open data
cyber, social, physical LOD on web
Barnaghi, P., Wang, W.: Semantics for the Internet of Things: early progress and
back to the future. International Journal on Semantic Web and Information Systems (2012)
http://thing.io/1
is a
http://ont/weather_sensor
CLASS
produces
http://thing.io/obs/1
http://ont/temp_observation
is a
13.0
has value
CLASS
℃
unit
located at
http://thing.io/loc/1
latitude longitude
-1.4150.9

6. LINKEDDATA ON
DISTRIBUTEDLIGHTWEIGHT COMPUTERS
Tiered levels of applications between the “Ground”
and “Cloud” for Meteorological Data
Irrigation Application
Query on
Rainfall
Meteorological Station on
Lightweight Computer
Data Stream
Environmental
Sensors
GROUND
Weather
Data
State Inclement
Weather Planning
Application
SRBench:National Weather and Disaster
Monitoring
CLOUD Broadcast Queries

7. PERFORMANCE
STORES DON’T SCALE & PERFORM WELLON WEB YET
Buil-Aranda, C., Hogan, A.: SPARQL Web-Querying Infrastructure: Ready for Action?
ISWC 2013
CHALLENGES
RESOURCE CONSTRAINED DEVICES
~500mhz CPU,512mb ram,SD CARD
historical and streaming data
Need storage andqueryefficiency ON BOTH

8. THE SHAPE OFIOT TIME-SERIES DATA
{
timestamp : 1467673132,
temperature : {
max: 22.0,
min: 15.0,
current: 17.0,
error: {
percentage: 5.0
}
}
}
FLAT
{
timestamp : 1467673132,
temperature : 32.0,
wind_speed : 10.5,
pressure : 1016
}
COMPLEX
20kUNIQUE DEVICES
dweet.io
99.5%FLAT SCHEMATA
0.5%COMPLEX SCHEMATA
1
2,3
4
5
6+
Width
{
timestamp : 1467673132,
temperature : 32.0,
humidity : 10.5,
pressure : 1016,
light: 120.0,
}
1
2
3
4

9. EFFICIENTQUERIESWITH
TIME-SERIES
DATA
THING
TEMPERATURE OBS
HUMIDITY OBS
WIND SPEED OBS
13.0
2016-01-0106:00:00
CELCIUS
93.0
2016-01-0106:00:00
PERCENT
10.5
2016-01-0106:00:00
MPH
LOCATION
produces
produces
located
produces
has value
unit
time
RDF
GRAPH

10. OBSERVATION DATA
OBSERVATION METADATA
SENSOR METADATA
THING
TEMPERATURE OBS
HUMIDITY OBS
WIND SPEED OBS
13.0
LOCATION
produces
produces
located
produces
has value
THING
THING
THING
TEMPERATURE OBS
timeTEMPERATURE OBS 2016-01-0106:00:00
unitTEMPERATURE OBS celcius
93.0has valueHUMIDITY OBS
timeHUMIDITY OBS 2016-01-0106:00:00
unitHUMIDITY OBS PERCENT
10.5has valueWIND SPEED OBS
timeWIND SPEED OBS 2016-01-0106:00:00
unitWIND SPEED OBS MPH
EFFICIENTQUERIESWITH
TIME-SERIES
DATA
RDF
TRIPLES

11. OUR
APPROACH
EFFICIENTQUERIESWITH
TIME-SERIES
DATA
THING
TEMPERATURE OBS WIND SPEED OBS
CELCIUS PERCENT MPH
LOCATION
produces
located
HUMIDITY OBS
unit
TEMPERATURE HUMIDITY WIND SPEED
13.0 93.0 10.5
TIME
2016-01-01 06:00:00

12. SENSOR METADATA
OBSERVATION DATA
OUR
APPROACH
EFFICIENTQUERIESWITH
TIME-SERIES
DATA
THING
TEMPERATURE OBS WIND SPEED OBS
CELCIUS PERCENT MPH
LOCATION
produces
located
HUMIDITY OBS
unit
TEMPERATURE HUMIDITY WIND SPEED
13.0 93.0 10.5
TIME
2016-01-01 06:00:00
OBSERVATION METADATA

13. DESIGNING OURENGINE
THING
TEMPERATURE OBS WIND SPEED OBS
CELCIUS PERCENT MPH
LOCATION
produces
located
HUMIDITY OBS
unit
TEMPERATURE HUMIDITY WINDSPEED
13.0 93.0 10.5
TIME
2016-01-01 06:00:00
Table1
TABLE1.TEMPERATURE
has value has value
TABLE1.HUMIDITY
has value
TABLE1.WINDSPEED

14. DESIGNING OURENGINE
THING
TEMPERATURE OBS WIND SPEED OBS
CELCIUS PERCENT MPH
LOCATION
produces
located
HUMIDITY OBS
unit
TEMPERATURE HUMIDITY WINDSPEED
13.0 93.0 10.5
TIME
2016-01-01 06:00:00
Table1
TABLE1.TEMPERATURE
has value has value
TABLE1.HUMIDITY
has value
TABLE1.WINDSPEED

15. DESIGNING OURENGINE
THING
TEMPERATURE OBS
CELCIUS PERCENT
produces
loc
HUMIDITY OBS
unit
TEMPERATURE HUMID
13.0 93.0
TIME
2016-01-01 06:00:00
TABLE1.TEMPERATURE
has value has va
TABLE1.H
MAX( )?TEMPERATURESELECT
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom
{
}
𝞹
𝞬 (max ( ))?TEMPERATURE
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom BGP

16. DESIGNING OURENGINE
TEMPERATURE OBS
CELCIUS
TEMPERATURE
13.0
TABLE1.TEMPERATURE
has value
MAX( )?TEMPERATURESELECT
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom
{
}
𝞹
𝞬 (max ( ))?TEMPERATURE
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom BGP

17. DESIGNING OURENGINE
MAX( )?TEMPERATURESELECT
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom
{
}
𝞹
MAX( )?TEMPERATURE
?OBS ?TEMPERATURE ?uom
TABLE1.TEMPERATURE CELCIUSNODE_TEMP
𝞹
𝞬 (max ( ))?TEMPERATURE
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom BGP

18. DESIGNING OURENGINE
MAX( )?TEMPERATURESELECT
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom
{
}
MAX( )TEMPERATURE
𝞹
𝞬 (max ( ))?TEMPERATURE
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom BGP
FROM TABLE1
𝞹

19. SPARQL
DESIGNING OURENGINE
MAX( )?TEMPERATURESELECT
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom
{
}
SQL SELECT MAX( )TEMPERATURE FROM TABLE1

20. LOWVISIBILITYSTATIONSEXAMPLE
?OBS VISIBILITY OBSa
has value?OBS ?VISIBILITY
has unit?OBS ?uom
}
{
FILTER( ,<10 )?VISIBILITY
UNION
?OBS RAINFALLOBSa
has value?OBS ?RAINFALL
has unit?OBS ?uom
}
{
FILTER( ,>30 )?RAINFALL
?SENSORSELECT {
}
UNION
?OBS SNOWFALLOBSa
has value?OBS ?SNOWFALL
has unit?OBS ?uom
}
{
𝞹
?OBS SNOWFALLOBSa
has value?OBS ?SNOWFALL
has unit?OBS ?uom BGP
UNION
UNION
FILTER( ,<10 )?VISIBILITY FILTER( ,>30 )?RAINFALL
?OBS VISIBILITY OBSa
has value?OBS ?VISIBILITY
has unit?OBS ?uom BGP
?OBS RAINFALLOBSa
has value?OBS ?RAINFALL
has unit?OBS ?uom BGP
FILTER( ,TRUE)?SNOWFALL

21. LOWVISIBILITYSTATIONSEXAMPLE
?OBS VISIBILITY OBSa
has value?OBS ?VISIBILITY
has unit?OBS ?uom
}
{
FILTER( ,<10 )?VISIBILITY
UNION
?OBS RAINFALLOBSa
has value?OBS ?RAINFALL
has unit?OBS ?uom
}
{
FILTER( ,>30 )?RAINFALL
?SENSORSELECT
}
UNION
?OBS SNOWFALLOBSa
has value?OBS ?SNOWFALL
has unit?OBS ?uom
}
{
𝞹
UNION
UNION
FILTER( ,<10 )?VISIBILITY FILTER( ,>30 )?RAINFALL
?OBS ?snowfall ?uom
TABLE1.snowfall BOOLNODE_SNOW
?OBS ?VISIBILITY ?uom
TABLE1.VISIBILITY cmNODE_vis
?OBS ?rainfall ?uom
TABLE1.rainfall cmNODE_RAIN
{
FILTER( ,TRUE)?SNOWFALL

22. LOWVISIBILITYSTATIONSEXAMPLE
?OBS VISIBILITY OBSa
has value?OBS ?VISIBILITY
has unit?OBS ?uom
}
{
FILTER( ,<10 )?VISIBILITY
UNION
?OBS RAINFALLOBSa
has value?OBS ?RAINFALL
has unit?OBS ?uom
}
{
FILTER( ,>30 )?RAINFALL
?SENSORSELECT {
}
UNION
?OBS SNOWFALLOBSa
has value?OBS ?SNOWFALL
has unit?OBS ?uom
}
{
𝞹
UNION
UNION
FILTER( ,<10 )VISIBILITY FILTER( ,>30 )RAINFALL
FROM TABLE1FROM TABLE1
FROM TABLE1
FILTER( ,TRUE)SNOWFALL

23. LOWVISIBILITYSTATIONSEXAMPLE
?OBS VISIBILITY OBSa
has value?OBS ?VISIBILITY
has unit?OBS ?uom
}
{
FILTER( ,<10 )?VISIBILITY
UNION
?OBS RAINFALLOBSa
has value?OBS ?RAINFALL
has unit?OBS ?uom
}
{
FILTER( ,>30 )?RAINFALL
?SENSORSELECT {
}
UNION
?OBS SNOWFALLOBSa
has value?OBS ?SNOWFALL
has unit?OBS ?uom
}
{
𝞹
UNION
FROM TABLE1
WHERE RAINFALL>30
SELECT SENSOR FROM TABLE1
WHERE VISIBILITY<10
UNION
SELECT SENSOR FROM TABLE1
FILTER( ,TRUE)SNOWFALL

24. SQL
SPARQL
LOWVISIBILITYSTATIONSEXAMPLE
?OBS VISIBILITY OBSa
has value?OBS ?VISIBILITY
has unit?OBS ?uom
}
{
FILTER( ,<10 )?VISIBILITY
UNION
?OBS RAINFALLOBSa
has value?OBS ?RAINFALL
has unit?OBS ?uom
}
{
FILTER( ,>30 )?RAINFALL
?SENSORSELECT {
}
UNION
?OBS SNOWFALLOBSa
has value?OBS ?SNOWFALL
has unit?OBS ?uom
}
{
SELECT
UNION
WHERE RAINFALL>30
WHERE VISIBILITY<10
UNION
SENSOR FROM
SELECT SENSOR FROM TABLE1
WHERE SNOWFALL==TRUE
(
)
SELECT SENSOR FROM TABLE1
SELECT SENSOR FROM TABLE1

25. HOWMAPPINGSWORK
SPARQL2SQL
Mapping
S2SML
MAPPING
TABLE1.TEMPERATURE
has value
THING
TEMPERATURE OBS WIND SPEED OBS
CELCIUS PERCENT MPH
LOCATION
produces
located
HUMIDITY OBS
unit
TABLE1.humidity
has value
TABLE1.WINDSPEED
has value
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom BGP
SPARQL
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom
SELECT * WHERE {
}
JENA
SESAME
??
INTERFACE

26. S2SML
MAPPING
WRITTEN IN RDF
REFLECTSTHE RDF ITREPRESENTS DIRECTLY
SUPPORTS COLLAPSING NODES
EFFICIENT QUERIES WITHIOT TIME-SERIES DATA
R2rML COMPATIBLE
CAN BE TRANSLATEDTOAND FROM R2RML
SUBJECT
object
predicate
IRImapIRI BNODE
IRI
FNODE
IRImapIRI BNODE FNODE
LITERAL LITERALMAP

27. S2SML
MAPPING
FauxNODE HUMIDITYOBS
TABLE1.COL1
hasValue
LITERALMAP
<http://knoesis.wright.edu/ssw/obs/{readings.uuid}>
if (HUMIDITY OBS is projected):
if(readings.uuid !exists):
create_col(readings.uuid)
return IRImap(readings.uuid)

28. RDF STREAMPROCESSING
sparql2stream
Same engine and
mappings but translates
to EPL instead of SQL
TRANSLATE
QUERY
2
Stream Window
SPARQL query specifying
stream window size
REGISTER
QUERY
1
Stream Sockets
Supports multiple
platforms and streams
with ZeroMQ
STREAMDATA
3
Real-time analytics
PIOTRe & Smart Home
Freeboard demo
github.com/eugenesiow/iotwo
RECEIVE PUSH
RESULTS
4

29. SPARQL2stream
RSP
QUERY
MAX( )?TEMPERATURESELECT
?OBS TEMPERATURE OBSa
has value?OBS ?TEMPERATURE
has unit?OBS ?uom
{
}
FROM NamedStream Obsstream Range 1h
EPL SELECT MAX( )TEMPERATURE
FROM Weather.win:time(1hour)
Event Processing
Language

30. EVALUATIONWITH BENCHMARKS
SRBENCH
~20,000 Stations
100 – 300k triples
Wind, Rainfall, etc.
10 SRBench Queries
Zhang, Y, et al. (2012) "SRBench: a streaming RDF/SPARQL
benchmark.”The 11th International Semantic Web Conference.
SMART HOME BENCH
Siow, E., Tiropanis, T., Hall, W. (2016). "Interoperable and Efficient:
Linked Data for the Internet of Things." The 3rd International
Conference on Internet Science.
3 months, 1 home
~30k triples
Motion, energy, environment
4 Analytics Queries
GraphDB (OWLIM)
Ontop
Our Approach (S2S)
TDB
G
Morph
O
S
M
T
Raspberry Pi 2 Model B+
1GB RAM, 900MHz Quad Core ARM
Cortex A7, Class 10 SD Cards

31. STORAGESIZE
3ook
HurricaneIke
1ook
NEVADABLIZZARD
3ok
SMARTHOME
OUR APPROACH(s2S)
TDB
x15
x68
x112
GraphDB x9
x1352
x453

32. Get the rainfall observed in a particular
hour from all stations01
02
SRBENCH QUERYRESULTS
Q01 with an optional clause
on unit of measure
x5
S2S
S
TDB GraphDB
Ontop Morph
x3
x13
x4k
x2
x4
x4
x5k

33. 03
04
05
Detect if a hurricane has been observed
Get the average wind speed at the stations
where the air temperature is >32
Join between wind observation and temperature
observation subtrees time-consuming in low resource
environment (Raspberry Pi)
Detect if a station is observing a blizzard
x3
x6
x6
x88
x3
x3

34. 06
07
08
Get the stations with extremely low visibility
Detect stations that are recently broken
Get the daily minimal and maximal air
temperature observed by the sensor at a
given location
x2
x14
x4
x6
x6
x5
x2

35. 09
10
Get the daily average wind force and direction
observed by the sensor at a given location
Get the locations where a heavy snowfall has
been observed
Our Approach (s2s) is shown to be faster on all queries
in the Distributed Meteorological System with SRBench
Join between wind force and wind direction observation
subtrees is time-consuming in low resource
environment (Raspberry Pi)
x3
x3k
x2
x7

36. Temperature aggregated by hour on a
specified day01
02
SMARTHOME RESULTS
Minimum and maximum temperature
each day for a particular month
S2S TDB GraphDB
x7
x29
x3
x9

37. 03
04
Energy Usage Per Room By Day
Diagnose unattended appliances consuming
energy with no motion in room
Our Approach (s2s) is shown, once again, to be faster on
all queries for Smart Home Analytics
Involves motion and meter data (much larger set), with
space-time aggregations and joins between motion and
meter tables/subgraphs.
Involves meter data (larger set), with space-time
aggregations.
x69
x13
x4

38. STREAMPROCESSING EFFICIENCY
1 2 3 4 5 7 8 9 10
SRBENCH
294 261
306
277k 3243k 5245
426
280k
98
Le-Phuoc, D., et al. (2011) "A native and adaptive approach for unified processing of
linked streams and linked data.” The 10th International Semantic Web Conference.CQELSPerformance Improvement
For IoT Data Over
SMART HOME
196
21
167
xImprovement
Query

39. STREAMPROCESSING SCALABILITY
VELOCITY
>99% <1ms latency increasing from 1 to 1000 rows/ms
VOLUME
33.5million rows, projected ~2.5 billion triples!
<1ms 10-100ms
1
2
5
10
100
1000
99% 100%
Rateinrows/ms
Percentage Latency in ms Bands

40. PERSONAL IOT REPOSITORY
Siow, E., Tiropanis, T. and Hall, W. (2016) PIOTRe: Personal Internet of Things Repository: The 15th International Semantic Web Conference P&D
github.com/eugenesiow/piotresparql2streamsparql2sql github.com/eugenesiow/sparql2sql
PIOTRE
Apps
sparql2stream
sparql2sql
Metadata

41. “It's a long road, it's a long and narrow way. If I can't work up to you,
you'll surely have to work down to me someday.”
Narrow Way by Bob Dylan
eugene_siow
Travel Supported By:
LINKED DATA FORINTEROPERABILITY
A rich model todescribe thingsand integrateconnected thing’sdata
OPTIMISED PERFORMANCEFORSTREAMS AND HISTORICAL DATA
FROM 2to3orders of magnitudeperformance improvement