Transcript ppt

On Data Management
in Pervasive Computing Environments
written by
Filip Perich, Anupam Joshi, Timothy Finin, and Yelena Yesha
Published in
IEEE Transactions on Knowledge and Data Engineering Vol. 16 No. 5 May 2004
Summerized By Sungchan Park @ IDS Lab.
2008-11-12
Overview
 The authors propose

A data management framework

For data-intensive, pervasive computing environments
–
Ad-hoc network environments
–
(Context-aware)Proactive caching
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 2
Challenge in Pervasive Computing
 Infrastrucure-based wireless network is not suitable for dataintensive pervasive computing

May lead to congestion in the wireless network and a bottleneck on
the yellow page hardware
Congestion!
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 3
Challenge in Pervasive Computing, cont.
 Thus, decentralized scheme(Ad hoc network) is required.

However, it is highly dynamic
–

Unstable, continuously changing…
So we need new data management method addressing the
dynamicity!
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 4
Data Management Challenge in Ad hoc Network
 Dynamicity in Ad hoc network
1.
Spatio-temporal variation of data and data source availability
2.
Lack of global catalog and schema
3.
No guarantee of reconnection
4.
No guarantee of collaboration
–

Some devices can refuse collaboration
Query answering is highly serendipitous!

We want to avoid such a situation

Each devices should gather information procatively!
–
predicting users’ future request.
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 5
Motivating Scenario
 Scenario

Bob made an appointment with Alice meeting at the mall. He input
the appointment into his mobile device.

He arrived the mall early so he shopped at the mall for a while.
Upon Alice’s arrival, Bob asks his mobile device to suggest available
restaurant in the mall.
His device cached such information during the mall exploration
predicting his future needs from his schedule.
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 6
MoTAGU
 Data management framework for proactive caching

It abstracts devices in terms of
–
Information Provider
–
Information Consumer
–
Information Manager: InforMa
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 7
MoGATU: elements
 Information Provider

Something can return data for request
–
Databases, …

Its capability is described in DAML+OIL

Registered in local InforMa
–
Reg=(s, p, I, t, a)

s : service model

p : process model(?)

I : input restrictions

t : lifetime of info

a : willingness level of collaboration
–
Communicate with only local InforMa
–
And local InforMa advertise local providers to other devices in vicinity
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 8
MoGATU: elements, cont.


Information Consumer

Something can query and update data

Not advertised

Send query to local InforMa
Information Manager

Maintain inforamtion
–
Local elements
–
Peer in vicinity
–


ID of devices

Types of information they can provide
Cache
Advertise local information provider
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 9
MoGATU: Big Picture
I have A!
I (may)need A!
And I know
you have one!
O.K.
Here are some A!
: Device
: Information Provider
: Information Consumer
: InforMa
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 10
MoGATU: Query Processing
 Annotated query

Query = (s, i)
–
–
s : service model

Requested data type

DAML+OIL
i : input value

Requested value
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 11
MoTAGU: Proactive Caching
 Creating standing queries for caching

Using user profile

Using current context

InforMa contains simple rules to create queries
–
Ex. “when user is driving in a car with low on gas, query to search gas
station”
 Cache replacement predicting needs

Preallocate some portion cache space for (maybe required) types of
data.

Using utility functions specified in user profiles to score cache.
–
Not time stamp based like LRU
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 12
MoTaGU: Routing
 Routing

Modifies AODV alg.

Best-effort basis
–
Attempts to rebuild
disconnected routes,
but do not guarantee
message delivery
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 13
Performance Experiments #1
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 14
Performance Experiments #2
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 15
Performance Experiments #3
 Cost on Reasoning

For 30 KB cache, 5ms per query after 100 runs
–
4.56s for communication


27ms at faster device
Reasoning is not dominant factor
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 16
Interesting Points
 Another usage of context

Using context for low level operation.
 Give logical explanation on “Why we should use ad hoc network
for pervasive computing?”
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 17
But I have some question…
 Q1: Is reasoning really not a big problem?

“find resturant” from “appointment” is quite high level intelligence!

Will this framework really work well on not made-up environment?
 Q2: Query for caching may lead to redundant communications,
and power can be consumed more because of this approach.
This point is not discussed properly.

And machine intelligence is not perfect!

With incorrect decisions, it is just a waste of power and network
resources.
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 18
Question Continues…
 Q3: Context awaring with info in single device is severly limited.
Can it really aware context at useful level?

Issuing additional query for building cache must not be allowed!

“location” and “time” may be enough for useful service.
 Q4: The authors’ argument that “we cannot build data-intensive
pervasive computing” looks quite logical. But is it really true?

If this is true, mobile service can not use information on large
database on TCP/IP network. it limits capabilities of pervasive
computing severly.

and devices not in vicinity also can make usefule collaboration!

It needs more discrete study on this issue.
Center for E-Business Technology
Copyright  2008 by CEBT
IDS Lab. Seminar - 19