Computer Science Technical Report Archive

USC Computer Science Technical Reports, no. 643 (1996)

(USC DC Other)

USC Computer Science Technical Reports, no. 643 (1996)

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Extending the Remote Exc hange framew ork to suit
instancebased in terop erabilit y
Mark V ermeer
Decem ber Abstract
In this do cumen t I discuss the Remote Exc hange pro ject an imp ort
exp ortbased sharing mec hanism implemen ting a database federation
Based on myo wn researc h I prop ose some extensions to this mec ha
nism and indicate ho w they migh t b e implemen ted
In tro duction
During m y sta y at the Univ ersit y of Southern California USC I ha v e
studied the Remote Exc hange RE pro ject that has been carried out at
USC during the last few y ears This do cumen t serv es to accoun t for the
w ork that I ha v e done during this sta y It describ es the RE pro ject and its
corresp ondences with myo wn w ork Based on ideas I dev elop ed I prop ose
some extensions to RE and describ e ho w they could b e implemen ted
This do cumen t is organised as follo ws I rst giv e a description of the
principal features of the RE system in Section In Section I then describ e
the idea of instancebased in terop eration whic h stems from m y earlier w ork
Section describ es ho w this paradigm relates to the curren t RE function
alit y Section con tains some prop osals for extensions of RE in the lightof
these ideas
An o v erview of Remote Exc hange
In this section I will describ e the main features of RE as I see it My sources
are publications ab out the pro ject suc h as
General principles
RE is a system based on the federated approac h to database in terop er
abilit y prop osed in That is unlik e man y other approac hes RE do es
not dene some in tegrated in terop erable lev el spanning all or a n um ber of
comp onen t databases Instead RE allo ws a comp onenttoimport dataex p orted b y other comp onen ts in to the lo cal database stressing the autonom y
of comp onen ts Th us eac h comp onen t is enric hed with imp orted data in a
transparentw a y without directly aecting the lo cal user in terfaces or query
pro cessing mec hanisms
An imp ortan t consequence of this is that instead of dening an in tegrated
sc hema that subsumes all lo cal sc hemata eac h comp onen t transforms remote
data to t its lo cal con text This mayin v olv e extensions to the lo cal sc hema
but conicts are alw a ys resolv ed in fa v our of the lo cal represen tation Also
a comp onen t will only imp ort remote information that is deemed relev an t
The whole pro ject consists of three tiers
The disc overy and iden tication of remote information relev an t to a
giv en comp onen t
The resolution of semantic heter o geneity bet w een this comp onentand
the exp orter of relev an t information
The actual mec hanism that enables these comp onen ts to shar e infor
mation
Disco v ery and Seman tic Heterogeneit y Resolution
T o imp ort ob jects from a remote source in a meaningful w a y information on
the me aning of remote ob jects is needed so that an imp orting comp onen t
ma y determine whether information pro vided b y a remote data source is
relev an t to him Moreo v er there ma y exist discrepancies in the w a y simi
lar information is represen ted in the imp orting resp exp orting comp onen t
Hence on top of the actual sharing mec hanism w e need functionalit y to
disc over relev an t remote information and to resolv e p ossible semantic het
er o geneity This functionalit y is part of the rst t w o tiers of RE listed ab o v e
These are discussed in this subsection
The main idea underlying the RE approac h to these issues is that one
should not reason ab out the meaning and resem blance of heterogeneous
ob jects in terms of their structural represen tation in the individual comp o
nen ts ie sc hematic information but use seman tic kno wledge ab out these

ob jects instead In particular a Semantic Dictionary SD is main tained
that serv es as a federated kno wledge base of the application domain co v ered
b y the federation comp onen ts This SD is dynamic in that it ev olv es with
the registration of new exp orted information The SD consists of c onc ept
hier ar chies where a concept is dened as a collection of similar t yp es
The actual pro cess of disco v ery and unication in RE is go v erned b y the
Sharing A dvisor SA a comp onen t that manages kno wledge ab out t yp e
ob jects that comp onen ts exp ort Note that the SA is primarily concerned
with t yp elev el sharing F our SA services are distinguished
R e gistr ation
When registering new t yp e ob jects to b e exp orted in to the SD a log
ical connection is established bet w een the exp orted information and
the existing concepts in the SD The determination of the relationships
bet w een the newly registered t yp e ob jects and the concepts in the cur
ren t hierarc h y is done using user in teraction supp orted b y a clustering
algorithm based on interc onc ept dissimilarity and intr ac onc ept sim
ilarity of prop erties to determine the meaning of a registered t yp e
ob ject
Disc overy
The disco v ery service tak es user comp onen t c haracteristics in to ac
coun t when determining relev an t remote information That is the
follo wing three t yp es of requests are supp orted
a Similar c onc epts This is a request for t yp es that are in the same
p ortion of the concept hierarc hyas a giv en lo cal t yp e
b Complementary information This is a request for a t yp e that
has additional information ie dieren t prop erties for a giv en
concept to whic h a certain lo cal t yp e b elongs
c R elate d information This is a request for a t yp e that shares
some of the prop erties of the concept to whicha giv en lo cal t yp e
b elongs This is useful for example for the sharing of beha viour
bet w een the lo cal and remote t yp e

Semantic Heter o geneity R esolution
The basic problem addressed b y the seman tic heterogeneit y mec hanism
is the determination of the relationship b et w een a lo cal and a remote
t yp e ob ject T o this end structur al kno wledge ab out t yp e ob jects
is represen ted b y metafunctions whic h are used to supply metadata
ab out remote t yp e ob jects suc h as their functions instances and sub
t yp es
Moreo v er a L o c al L exic on LL represen ts the semantic relationships
bet w een lo cal t yp es and globally understo o d concepts from the SD
P ossible relationships include iden tit y ie same concept same repre
sen tation equiv alence same concept dieren t represen tation com
patibilit y kind of asso ciation eg synon yms t ypically used in the
same con text collection of has a prop ert y that is inheren t to the
t yp e and descriptiv e feature of Hence the LL represen ts the seman
tics of shared terms in a more expressiv e and complete manner than
in the conceptual sc hema
Th us in RE the meaning of concepts is appro ximated using a
com bination of the SD whic h con tains partial information on in ter
comp onen t concept relationships and the LL whic h is an attempt to
describ e the precise meaning of lo cal t yp es in terms of globally under
sto o d concepts
Unic ation
When folding relev an t remote information in to the lo cal database se
man tic discrepancies b et w een the lo cal and remote comp onentm ust b e
dealt with F or example the v alues of primitiv e stored functions ma y
be c onverte d to some common domain Moreo v er functions andor
t yp es maybe r ename d to deal with homon yms and synon yms
When the seman tic relationship bet w een a remote t yp e and a lo cal
t yp e has b een determined this t yp e can b e added to the lo cal sc hema
The w a y in whic h this is done dep ends on the relationship in v olv ed
As w e touc h on the actual sharing mec hanism here this is discussed
in the next subsection
The sharing mec hanism
Three sorts of ob jects to b e shared are distinguished

Instance ob jects
The imp ort of remote instance ob jects can b e seen as the in tegration
of remote information in to the lo cal con text Lo cally surr o gate obje cts
are created to represen t the imp orted ob jects The functions asso ci
ated with the lo cal t yp e under whic h the remote ob ject is imp orted
are overridden to do an RPCcallofthe appropriate remote function
asso ciated with the original remote ob ject this is done using multi
ple typing for surrogate ob jects ie surrogate ob jects are both of a
surrogate remote t yp e and of the t yp e they are imp orted in to and
corresp onding dynamic binding of o v erloaded functions
T yp e ob jects
The imp ort of remote t yp e ob jects can be seen as the em b edding of
a remote con text in to the lo cal con text A remote t yp e is alw a ys im
p orted as a subtyp e of a lo cal t yp e and subt yp es of the imp orted t yp e
are automatically imp orted as w ell T ypically imp orting a remote
t yp e in v olv es the sharing of b oth the metadata ob jects and the re
mote instances that form its extension Subsequen tly additional lo cal
instances ma y b e added to these remote t yp es
When imp orting a single remote t yp e R in to a lo cal comp onen t t w o
t yp es of relationships b et w een R and a lo cal t yp e L are distinguished
a R is seman tically equiv alentto L No w R is lo cally created as a subt yp e of L and all its functions
are created for b oth R and L b R is related to L
No w the w ellkno wn generalisation tec hnique is used both R and
a new sup ert yp e S are created where S is a sup ert yp e of b oth R
and L S con tains the common functions of R and L If there is no suc h relationship R is simply imp orted as the ro ot of a
new hierarc h y If multiple interr elate d remote t yp es are imp orted rst the t yp e ob
jects are imp orted as sk etc hed ab o v e follo w ed b y the functions that
relate them Finally in the presence of r epr esentation dier enc es of re
lationships b et w een equiv alen t lo cal and remote t yp es a surrogate for
the lo cal relationship represen tation is created for eac h pair of related
imp orted ob jects

Beha vioural ob jects
These ob jects can be either stored functions often called attributes
in other data mo dels or computed functions whic h corresp ond to
obje ct r etrieval metho ds By sharing beha viour a comp onen t ma y
access remote services for its ob jects that are not supp orted lo cally Sharing of b eha viour can b e done in either of the t w o follo wing w a ys
Implicit beha viour sharing is p erformed when imp orting remote in
stances its asso ciated functions are shared automatically Explicit
b eha viour sharing is done b y rst imp orting remote function ob jects
and then asso ciating them with lo cal t yp es A c al lb ack me chanism is
supplied for shared remote functions to access the state of lo cal ob jects
they are executed on
Hence in con trast to most existing approac hes the unit of sharing b et w een
comp onen ts is not necessarily a c ol le ction of ob jects or in particular the
complete extension of a certain remote t yp e instance ob jects are an appro
priate unit of sharing here and ev en b eha vioural ob jects functions
A b eha viour sharing taxonom y An imp ortan t feature of RE is the
decoupling of lo cation of data and b eha viour Hence wema y execute
Lo cal functions on lo cal ob jects
This is the base case regardless of whether the function is stored or
computed
Lo cal functions on remote ob jects
This has the eect of giving lo cal state stored functions or lo cal b e
ha viour computed functions to a remote ob ject Th us these ob jects
are adapted to the lo cal con text
Remote functions on lo cal ob jects
This only mak es sense if the function is computed in that case the
remote database is pro viding a service to lo cal ob jects that cannot b e
supplied lo cally This ma y be seen as the purest form of beha viour
sharing
Remote functions on remote ob jects
This is the basis of the instance sharing mec hanism as poin ted out
ab o v e b eha viour sharing is implicit in this case

Instancebased database in terop eration
It ma y b e observ ed that there exists a certain gap b et w een the t w o principal
comp onen ts of RE the sharing advisor and the actual sharing mec hanism
Whereas the fo cus of the sharing advisor is on detecting relationships be t w een and the unication of lo cal and remote typ es the actual sharing mec h
anism is mostly concerned with the sharing of instances and beha viour A
p ossible w a y to bridge this gap is b y applying the principle of instancebased
database in terop eration In this section w e presen t the main ideas b ehind
this paradigm as presen ted in Ob jects and classications
Observ e that the notion of a sc hema is essen tially a classic ation of the set of
realw orld ob jects app earing in a certain application domain Ob jects that
are considered similar bythe sc hema designer are group ed in to entity typ es
classes and describ ed using a common set of prop erties attributes If the
set of realw orld ob jects group ed b y a certain en tit y t yp e E is a subset of
the set of realw orld ob jects group ed b y another en tityt yp e E
E is called
a subt yp e of E
W e note that the classication of realw orld ob jects in to en tityt yp es is
inheren tly subje ctivedieren t designers will do it in dieren tw a ys dep end
ing on their w a y of lo oking at the application domain Moreo v er due to
dierences in c ontext it is often v ery hard to determine the exact relation
ship b et w een a lo cally dened en tityt yp e and a remotely dened en tityt yp e
T ak e as a simple example the en tit y t yp e Person dened in the database
of the CSdepartmen tof univ ersit y X and the en tityt yp e Student dened
in a database for the whole univ ersit y X W e w ould argue that the ap
paren t ISArelationship bet w een these t yp es do es not hold here since the
more general t yp e is dened in a more sp ecic con text Also consider en tit y
t yp es lik e Federated DBS v ersus MultiDBS or ev en AmericanCars v ersus
BeautifulThings In short our idea is to a v oid ha ving to dene relationships bet w een
classes as m uc h as p ossible Instead w e try and dene relationships b et w een
instanc e obje cts W e then apply b oth the lo c al and the r emote classic ation
to an appr opriately mer ge d glob al set of obje cts
Sp ecifying in terop erabilit y
Our approac h requires a designer to sp ecify conditions under whic h a certain
relationship b et w een a remote ob ject O
and a lo cal ob ject O or class C holds
The relationships w e distinguish are
Identity O and O
represen t the same realw orld ob ject This is
represen ted as Eq O
O
Strict similarity O
w ould lo cally b e classied under C This is rep
resen ted as Sim O
C
Appr oximate similarity Lo cally C f O
g w ould b e considered a mean
ingful more general class C

This is represen ted as Sim O
CC
Descriptivity Lo cally O
is considered a set of v alues S describing an
ob ject O
whic h is iden tical to a lo cal ob ject O or similar to a lo cal
class C This is represen ted as Eq O
OS or Sim O
CS Constituency Lo cally O
is seen as a constituen t of O O and
O
describ e dieren t lev els of aggregation This is represen ted as
Aggr O O
W e require the sp ecication of obje ct c omp arison rules o crs of the form
where is an y of the relationships listed ab o v e and is a conjunc
tion of rstorder logic predicates whic h mightin v olv e additional informa
tion suc h as corresp ondence tables etc
Moreo v er pr op erty e quivalenc e assertions prop eqs ma ybeform ulated
for eac h pair of classes bet w een whic h o crs ha v e b een sp ecied sp ecifying
to what exten t the descriptions pro vided b y DB and DB
o v erlap This is
just what is done in regular sc hema in tegration These assertions are of the
form pr opeq Cp C
p
cf cf
d f where
p p
are basic or deriv ed lo cal and remote prop erties resp ectiv ely cf cf
are c onversion functions mapping the domains of p and p
to a
common domain Dand
d f D D D is a de cision function whic h determines a global
v alue for the prop ertygiv en p ossibly dieren t lo cal and remote v alues
W e require that for eac h decision function d f a D jd f a a a In our view functions suc h as sum used eg in dene deriv ed
global prop erties rather than determining v alues for equiv alen t lo cal
and remote prop erties

Publisher
Affil(): String
Name():String
Paper
Research
Book
IEEE ACM
AppIn: String()
Abstract(): String
PublishedBy(): Publisher
Sig(): String
NrPages(): Integer
Subjects(): P String
Name():String
Location(): String
Researcher A Researcher B
B1
B2 B3
B4
B5 B6
B7
Publ
Auth
Name(): String
Venue(): String
Chapter
Issue(): Integer
Vol(): Integer
Journal
Conference
Publication
Editor(): Author
Booktitle(): String
A3
A4 A5
A6
A7
Name(): String
Title:() String
Date(): Integer
Authors(): P Author
Title:() String
Author
Affil(): String
Name():String
A1 A2
TextBody(): String
Publisher(): String
View() USES TextBody
Date(): Integer
Authors(): P Author
MakeRef(Style): String USES Authors, Date
Figure Researc hers As and Bs sc hemata
Example
Figure depicts the sc hemata of the literature databases main tained b y
a researc her A and a researc her B resp ectiv ely Note that the databases
con tain similar information but there app ears to be some discrepancy in
the classications used F or example the term Publication as used b y
researc her A ma y v ery w ell be closer to what B calls a Researc h pap er
than a Publ Moreo v er A distinguishes sub classes based on the t yp e of
publication forum while B considers the organisation resp onsible for the
publication
Supp ose researc her B w ould lik e to imp ort some information from re
searcherAin to his o wn database He migh t then dene the follo wing ob ject
comparison rules
Eq O PublO
Publication O title O
title a O authorsa
O
authors ja name a
name
Sim O
ConferenceO IEEE contains O
name IEEE
Sim O
JournalO IEEE contains O
name IEEE
Sim O
ConferenceO ACM contains O
name A CM
Sim O
JournalO ACM contains O
name A CM
Eq O
Publication publisher O Publisher O
publisher O name
Sim O
Publication publisher O Publisher
Ag g r O
ChapterO Book O
booktitle O title
Eq O AuthO
Author O name O
name
Note that the uncertain t y ab out the exact relationship bet w een Publ and
Publication do es not prev en t us from dening iden tit y rules b et w een their
instances Strict similarityis more con v enien tly dened lo w er in the hierar
c h y where meaning b ecomes more ob vious ho w ev er
F unction names and domains ha v e been k ept similar lea ving only ob vi
ous prop ert y equiv alences except for the follo wing
pr opeq Publicationauthors Publauthors id id union Instancebased in terop eration and RE
The sharing mec hanism supplied b y RE seems to be an appropriate plat
form to demonstrate the use of instancebased in terop eration in the con text
of an imp ortexp ortbased database in terop eration arc hitecture Ob ject
comparison rules can be used as a sp ecication la y er on top of the shar
ing mec hanism pro viding conditions under whic h dieren t forms of sharing
are applied F urthermore the functionalit y pro vided b y the sharing advisor
ma y b e used to select candidate lo cal and remote classes b et w een whichob ject comparison rules are to b e sp ecied Th us sp ecications consisting of
ob ject comparison rules and prop ert y equiv alence assertions could be used
as an interfac e bet w een these RE comp onen ts
Instancebased in terop eration in an imp ortexp ort
paradigm
Adapting the idea of instancebased in terop eration to the imp ortexp ort
paradigm with its stress on lo cal autonom y the follo wing principles suggest
themselv es
Rely on instance sharing as m uc h as p ossible
Where p ossible adhere to the lo c al classication for the mixed set of
nativ e and imp orted instances th us obtained
In terms of our example of Figure researc her B w ould imp ort re
searc her As publications but classify them using his lo cal t yp es where

p ossible B apparen tly do es not care whether a publication is a con
ference pap er or a journal pap er As ob jects should b e classied ac
cording to Bs in terests whether they are published in A CM or IEEE
fora
Use remote t yp es as surrogate t yp es imp ort remote t yp es in to the lo cal
hierarc h y only to hold imp orted instances that cannot b e assigned to
anylocal t yp e
Use b eha viour sharing to access additional functions that the remote
t yp es supply for the shared instances
Note that these principles partly replace the curren t suggestions for t yp e
unication supplied b y the Sharing Advisor T o implemen t these ideas
w e m ust extend the curren t RE sharing mec hanism with some additional
functionalit y This is discussed in the next section
Ob ject relationships and surrogates
Although the idea of con trolling the imp ort of remote instances through
the denition of ob ject comparison rules seems an attractiv e one there is a
discrepancy b et w een the notion of conditions for ob ject relationships and the
surrogate mec hanism that implemen ts instance sharing in RE The surrogate
mec hanism is static in the sense that it represen ts the extension of remote
t yp es at a certain poin t in time The mec hanism do es not easily reect
c hanges to the extension of aremote t yp e F or example the remote ob ject
o
represen ted b y a surrogate so in the lo cal database migh t be deleted
from the remote database Although in principle this should result in a
corresp onding deletion of so from the lo cal database suc h an automatic
deletion violates lo cal autonom y Th us autonomyof componen ts causes the
instance sharing mec hanism to b e static
In the presence of conditions that go v ern the imp ort of remote instances
as in ob ject comparison rules this problem is expanded to c hanges in the
state of the original remote database F or example an up date to a remote
ob ject o
mayin v alidate the condition o
under whic h o
w as imp orted in to
the lo cal database as the surrogate so Alternativ ely suc h an up date ma y
lead to this condition to b e true on ob jects that did not qualify previously These problems are kno wn as the view maintenanc e pr oblem in essence
the surrogate mec hanism implemen ts a materialised view In view of the
autonom y requiremen t in RE I here assume that remote c hanges are not to
b e immediately reected in the lo cal database

Ob ject comparison rules and the Sharing Advisor
As I prop ose to let ob ject comparison rules to playthe role of an in terface
bet w een the Sharing Advisor and the sharing mec hanism the resp onses of
the SA to disco v ery requests are to b e dened as follo ws
In resp onse to a request for similar information the SA suggests re
mote t yp es for the denition of strict similarity rules along with sug
gested prop ert y equiv alences Suc h rules lead to the imp ortation of
additional instances to the lo cal t yp es
In resp onse to a request for c omplementary information the SA sug
gests remote t yp es on whic h to dene identity rules along with sug
gested prop ert y equiv alences Suc h rules cause the curren t instances
of the lo cal database to b ear additional information
In resp onse to a request for r elate d information the SA suggests re
mote t yp es on whichto dene appr oximate similarity rules along with
suggested prop ertyequiv alences Suc h rules add related foreign t yp es
and their instances to the lo cal database
When the SA detects structur al c onicts bet w een a lo cal t yp e and a
related remote t yp e it suggests to dene descriptivity rules bet w een
them
When the SA detects the COLLECTIONOF relationship b et w een a
lo cal t yp e and a related remote t yp e it suggests to dene c onstituency
rules b et w een them
Extending the RE sharing mec hanism Some
prop osals
In this section I describ e the extensions to the RE sharing mec hanism that
are necessary to implemen t the ideas ab o v e I discuss the implemen tation
of eac h of the ob ject relationships distinguished ab o v e and the sharing of
complex ob jects
Strict similarit y
Strict similarityis supported b y instance sharing Remote instances o
that
satisfy the condition o
under whic h they are dened to b e strictly sim

ilar to the lo cal t yp e T are imp orted in to that t yp e F or example all
Conference instances whose name con tains A CM are to b e imp orted in to
ACM Equiv alen t functions of o
and T are dealt with through the standard
RPC calls of the surrogate mec hanism W e do not assume that equiv alen t
lo cal and remote functions ha v e the same name the RPC mec hanism can
hide homon yms and synon yms Instead of imp orting a remote function as
yielding v alues in the remote domain of the function ho w ev er w e can adapt
it to lo cally adhered domains b y applying a con v ersion function cf
to its
result th us o v ercoming v alue incompatibilities
Note that imp orted remote ob jects ma y supp ort additional functions
that do not o ccur in t yp e T W e ha v e t w o options for dealing with suc h
functions
These functions ma y be collected in a new subt yp e S of T that is
added to the lo cal hierarc h y and con tains the surrogates o ccurring in
T do not confuse S with the surrogate t yp e for the imp orted ob jects
whic h is used b y the RPCo v erriding mec hanism and is in visible to
the lo cal user
Alternativ ely these functions ma y be added to the lo cal t yp e itself
through b eha viour sharing This is esp ecially attractiv e for computed
functions F or stored functions the lo cal user is then resp onsible for
pro viding v alues for these functions on the nativ e ob jects
Ob ject iden tit y
In the curren t v ersion of RE remote instances that are imp orted in to the
lo cal database are implicitly assumed to represen t a realw orld ob ject that
is not represen ted in the lo cal database y et In m y terminology the ob ject
is assumed to b e strictly similar to the lo cal class but not iden tical to an y
of the lo cal instances
Ho w ev er in terms of our example some of the publications con tained
in researc her As database ma y already exist in researc her Bs database
Using b eha viour sharing and b yin tro ducing some simple extensions to the
instance sharing mec hanism w e can allo w for the imp ortation of iden tical
ob jects as w ell
A surrogate t yp e is in tro duced for the iden tical remote ob jects The
t yp e is completely iden tical to the t yp e holding the surrogates for
the strictly similar imp orted instances except for the w a y duplicately
named stored functions are o v erridden as explained b elo w

Abstract(): String
PublishedBy(): Publisher
Title:() String
Date(): Integer
Authors(): P Author
Paper
Research
ACM
AppIn: String()
B5
Publ
R-Publication
R-OID(): String
FullACM
Sig(): String
Surrogate
Types
R-Functions
Name(): String
Vol(): Integer
Issue(): Integer
MakeRef(Style): String
MakeRef(Style): String
View()
TextBody(): String
Publisher(): String
Date(): Integer
Authors(): P Author
Title:() String
TextBody(): String
View()
USER VIEW
R-Journal-Sim
R-Journal-Eq
B6 S1
Figure Iden tit y and strict similarit y
The lo cal instance to whic h the imp orted instance is iden tical is added
as an instance of the new surrogate t yp e Note that w e need not create
a new surrogate here instead it m ust b e p ossible to dynamically assign
an additional t yp e to an instance
Ov erriding of lo cal functions is nowasfollo ws Instead of just calling
the remote function through an RPC as is done for strictly similar ob
jects the function is ev aluated as d f cf
RP C r f cf l ocf unc Th us
the lo cal and remote results of the function are resolv ed through the
decision function as sp ecied in the prop ert y equiv alence assertion
P ossible additional remote functions are dealt with as sk etc hed under
strict similarit y Example Iden tit y vs strict similarit y Again consider the example of
Figure Supp ose w e ha v e the follo wing ob ject relationships Eq A B
and Sim A ACM Determining suc h relationships in v olv es ev aluating the
conditions sp ecied in the o crs against the state of these ob jects F or ex
ample AName mightbeA CM T r on SE

Figure illustrates ho w the ob jects A and A are imp orted in to re
searc her Bs database Note that the lo cal stored functions Abstract Ap
pIn and Sig are initialised to for S whic h is the lo cal surrogate for
A
The example illustrates both of the approac hes to incorp orating addi
tional remote functions men tioned ab o v e While the computed Mak eRef
function is shared and added to the lo cal t yp e Publ the additional remote
functions T extBo dy and View are catered for using a new subt yp e of
ACM These functions are a v ailable for surrogate ob jects only Ob ject iden tit y and the b eha viour sharing taxonom y Note that this
mec hanism in tro duces a new typeofbeha viour sharing imp orting a remote
stored function to a lo cal instance the function T extBo dy on B in our
example Curren tly this t yp e of beha viour sharing is seen as somewhat
meaningless Weha v e just seen that this t yp e of sharing do es mak e sense if
there exists an equiv alen t of the lo cal instance in the remote database Hence
this t yp e of sharing is the basis for the imp ortation of remote instances that
are iden tical to some lo cal instance
Note that I discuss sharing of computed functions including the conse
quences of iden tit y bet w een lo cal and remote instances in a separate sub
section
Appro ximate similarit y
In con trast with strict similarit y and iden tit y the instances imp orted in the
case of appro ximate similarit y cannot b e assigned to at yp e in the lo cal hi
erarc h y Hence w e resort to t yp e sharing to supp ort appro ximate similarit y Supp ose w e ha v e an appro ximate similarit y rule bet w een the remote t yp e
T
and the lo cal t yp e T This can b e supp orted bythe sharing mec hanism
as follo ws
Create a new lo cal virtual sup ert yp e S of Tcon taining the equiv alen t
prop erties of T and T
S do es not ha vean y direct instances
Imp ort the remote t yp e T
as a subt yp e of S Imp ort all instances of T
that satisfy o
the condition under whic h
T
ob jects are appro ximately similar to T The dierence b et w een the computed functions View and Mak eRef is discussed in
the section on b eha viour sharing

Appro ximate similarit y do es not o ccur in our example it is straigh tforw ard
to implemen t
Structural dierences descriptivit y
A structural dierence or descriptivit y relationship bet w een comp onen ts
o ccurs when there is an equiv alence b et w een a set of functions eac h return
ing a simple t yp e o ccurring in one comp onen t and a function returning a
userdened t yp e in another comp onen t In the con text of the RE sharing
mec hanism w em ust distinguish t w o cases
The lo cal t yp e LT con tains the set of simplev alued functions the
imp orted remote t yp e RT con tains an equiv alen t function returning
the userdened t yp e RU T The solution here is to equip the surrogate t yp e of RT with simple
functions f as in the lo cal database ev aluated b y p erforming an RPC
that follo ws the corresp onding path RU T f in the remote database
Alocal t yp e LT con tains a function f returning the userdened t yp e
LU T the imp orted remote t yp e RT con tains an equiv alen t set of
simplev alued functions
The solution here is to create two surrogate ob jects for eac h imp orted
instance ob ject o RT one s to b e imp orted in to LT and one s to b e imp orted in to LU T Both obtain their v alues from executing an
RPC on the same remote instance Moreo v er f s def
s
Note that structural dierences cause us to c hange the onetoone relation
ship b et w een surrogates and remote ob jects in to a man ytoone relationship
Example Our example illustrates the second of the cases distinguished
ab o v e The lo cal t yp e Publisher do es not ha v e an equiv alen t in the re
mote database instead the remote t yp e Publication con tains a func
tion PublisherString The extended sharing mec hanism creates surrogate
Publication instances from the remote Publisherv alues Note that ac
cording to the ocr sp ecied suc h ob jects need to be created only if there
do es not exist a Publisherob ject with the same name already publishers
with the same name are assumed to b e the same In other w ords virtual re
mote VirtPublisher ob jects can ha v e iden tit y and similarit y relationships
with lo cal ob jects to o

Abstract(): String
PublishedBy(): Publisher
Title:() String
Date(): Integer
Authors(): P Author
USER VIEW
R-Publication
R-OID(): String
Surrogate
Types
Name(): String
Vol(): Integer
Issue(): Integer
R-Journal-Sim
R-Journal-Eq
R-Functions
MakeRef(Style): String
View()
TextBody(): String
Date(): Integer
Authors(): P Author
Title:() String
PublishedBy(): String
ACM
B5
Publ
FullACM
MakeRef(Style): String
TextBody(): String
View()
B6 S1
Paper
Research
Publisher
Name(): String
Location(): String
Sig(): String
B1
S2
AppIn: String()
S2 B1
B1
Publisher
R-Virt-
Virt-Surrogate
Types
Name(): String
Figure Descriptivit y
In our example w e assume that APublisher B Name hence a
surrogate need not b e created P ossiblyho w ev er this in tro duces a conict
in the PublishedBy v alue b et w een A and its lo cal coun terpart B whic h
migh t assign a dieren t lo cal ob ject as its publisher Suc h a conict is then
solv ed using the decision function for PublishedBy as sp ecied b y the
in terop eration designer The v alue of APublisher is assumed to giv erise
to a new surrogate Publisher ob ject S whic h of course returns for
Lo cation Note that surrogates S and S stem from the same remote
ob ject A Note furthermore that since the remote virtual ob jects do
not pro vide o v erlapping additional information for Publishers a distinction
bet w een strictly similar and iden tical RVirtPublisher ob jects need not
b e made
Sharing complex ob jects
Complex ob jects are made up of comp onen t ob jects with relationships
among them ob ject relationships are implemen ted b y functions whose re
turn t yp e is a nonprimitiv e one Complex ob jects are an essen tial part of

the ob jectorien ted approac h but man y OO in terop erabilit y framew orks
neglect them Although in the do cumen tation of the Sharing Advisor this
sub ject is addressed at the t yp e lev el at the lev el of the actual sharing
mec hanism there is no concept of complex ob ject sharing Here I discuss
ho w complex ob ject sharing could b e realised in the con text of RE
Imp orting related instances
Consider the general case of a complex ob ject relationship where a re
mote instance o of t yp e T has a stored function f
with t yp e setof T and v alue S fo o o n
g Sharing of suc h a complex ob ject in
v olv es sharing of the ob jects o o o o n
b y creating lo cal surro
gates so so so so n
resp ectiv ely Also the function f
m ust be
shared
Ho w this is done dep ends on t w o factors
Whether or not r elationship mer ging m ust b e p erformed Relationship
merging o ccurs if there exists a lo cal ob ject o whic h is iden tical to o with a relationship implemen ting function f whic h is equiv alentto f
f and f
m ust then b e merged for o Whether the relationship is shared in a materialise d or a virtual w a y Virtual vs materialised relationship sharing The latter t w o mo des
of relationship sharing are no w explained in the situation where relationship
merging do es not o ccur
Materialised relationship sharing refers to the situation where the func
tion f on so is dened as fso so so n
g That is the rela
tionship is dened as is at the momen t of imp ortation from the remote
source P ossible up dates to the v alue of f are no w under con trol of
the imp orting database c hanges to the v alue of f
are not reected in
the v alue of f Virtual relationship sharing occurs when f
is dened as Sur r RP C f
where RP C f
retriev es the v alue of
of f
from the remote database and Sur r fo
o
o
n
g returns the
set fso
so
so
n
g of lo cal surrogates for the remote ob jects The
implemen tation of Sur r S requires a query of all surrogate ob jects
whose ROID v alue o ccurs in S and the creation of new surrogates
for the remote ob jects for whic h a lo cal surrogate do es not exist y et

Th us with virtual relationship sharing f is lo cally implemen ted as a com
puted function while with materialised relationship sharing f is a lo cally
stored function
Relationship merging An in teresting situation arises when in ad
dition to the remote ob ject o ha ving a relationship with ob jects
fo o o n
g implemen ted in the function f
there exists a lo cal ob ject
o with a function f fo o o m
g where o is iden tical to o and
f is equiv alentto f
In order for this relationship to b e shared the decision
function d f asso ciated with the prop ertyequiv alence assertion for f and f
m ust b e tak en in to consideration Tw o options seem reasonable
d f union
Assuming virtual relationship sharing this redenes o f as o f
old
Sur r RP C o f
Note that the union op erands need not be dis
junctiv e some ob jects in o f
old
maybe iden tical to those in o f
In
other w ords ob jects are considered to b e related if they are related in
an y of the databases
d f inter sect
Assuming virtual relationship sharing this redenes o f as o f
old
Sur r RP C o f
Here the set of related ob jects are exactly those
ob jects of o f
old
and o f
that are iden tical In other w ords ob jects
are considered to b e related if they are related in b oth databases
Example Again consider our example of Figure Recall that w e assumed
the relationship Eq A B F urthermore assume AAuthors fA A g B Authors fB B g and the additional relationship Eq A B The imp ortation of Aauthors in to Bs database in v olv es the creation
of a surrogate S for A No surrogate needs to be created for A since
the iden tical ob ject B already exists in the lo cal database The result of
applying B Authors immediately after imp ortation is fB B S g if the
decision function is unionand fB g if the decision function is interse ct See
Figure Constituency
The nal ob ject relationship to b e discussed constituency is related to the
issue of complex ob ject sharing in the sense that constituency refers to a

A6 B6
A1 A2 B2 B3
Authors Authors
Eq
Eq
Authors
B6(A6)
B2(A2) S3(A3) B3
Figure Relationship merging
complexob ject relationship b et w een ob jects in dieren t comp onen ts That
is if there exists an ob ject comparison rule Ag g r O O
O
O bet w een
a lo cal ob ject O T and a remote ob ject O
T
O
is to b e imp orted as a
part of a complex ob ject ro oted at O This is implemen ted b y imp orting T
and its instances in to the lo cal hierarc h y and extending the t yp e T with a
computed function f T
with b o dy return Sur r fO
j O
O g In our running example the ob ject relationships Ag g r A B and
Ag g r A B whic h migh t be determined based on a corresp ondence be t w een their resp ectiv e Bo oktitle and Title v alues as indicated b y the
constituency rules w ould lead to the denition of a stored function Chap
ters on Book where B Chapters fS S g the lo cal surrogates for A and A The beha viour sharing mec hanism
Although this seems to be suggested b y the curren t RE pap ers not ev ery
b eha vioural ob ject can b e shared among comp onen ts One cannot apply an y
lo cal computed function to a remote ob ject or vice v ersa b ecause sucha
function t ypically requires access to a giv en set of attributes whic hmaynot
b e supplied for the remote lo cal ob ject it is applied to T o apply a function
f to ob jects of original t yp e T the set F
f
of functions called b y f should

b e a subset of the functions dened on T Example Referring to our example of Figure the function View sup
plied b y researc her A on his Publication ob jects cannot be applied to
researc her Bs Publ ob jects that ha veno equiv alen t Publication ob ject in
As database The View function calls the stored function T extBo dy on
the ob jects it is applied to and this function is not supplied b y Bs database
Since the imp ortation of iden tical ob jects from A to Bcan b e seen as sup
plying T extBo dy v alues for Bs ob jects View can be executed against
those Publ ob jects for whic h an iden tical Publication ob ject exists only This is reected in the t yp e hierarc h y of Figure in that View is dened
on the subt yp e FullACM only As Mak eRefSt yle function on the other hand can be applied to Bs
Publob jects since these pro vide all stored functions necessary to compute
Mak eSt yle Hence in Figure s t yp e hierarc hyMak eRefSt yle is imp orted
as a function on Publ The USES clause Hence the sp ecication of the b o dy of the shared func
tion in addition to just its signature as in TMlik e formalisms or at least
a sp ecication of the functions that are called b y a computed function is
called for In the example I ha v e added the USES clause to the declaration
of computed functions
The necessit y for shared computed functions to access the state of the
ob jects they are computed on is ac kno wledged b y the creation of the c al lb ack
mec hanism but w eneed tokno w on b eforehand whic h functions a callbac k
will b e made for and it m ust b e assured that these are indeed a v ailable
Conclusion
In this do cumen t I discussed p ossible extensions to the RE framew ork
whic h stem from m y o wn previous researc h It app ears that RE is a suit
able en vironmen t for instancebased database in terop eration and that its
sharing mec hanism can easily be extended to fully supp ort instancebased
database in terop eration sp ecications consisting of ob ject comparison rules
and prop ert y equiv alences Ob ject comparison rules th us act as an in terface
bet w een the Sharing Advisor and the sharing mec hanism

References
D F ang S Ghandeharizadeh and D McLeo d An exp erimen tal ob ject
based sharing system for net w ork ed databases The VLDB Journal pp J Hammer D McLeo d and A Si An in telligen t system for iden tifying
and in tegrating nonlo cal ob jects in federated database systems Univ er
sit y of Southern California Los Angeles CA J Hammer and D McLeo d An approac h to resolving seman tic hetero
geniet y in a federation of autonomous heterogeneous database systems
In ternational Journal of In telligen t and Co op erativ e Information Systems no D F ang J Hammer and D McLeo d A mec hanism and exp erimen tal
system for functionbased sharing in federated databases in IFIP In
terop erable Database Systems DS Lorne Victoria Australia
No v em b er Amsterdam NorthHolland pp
D Heim bigner and D McLeo d A federated arc hitecture for information
managemen t A CM T rans O Inf Syst no pp July
M W W V ermeer and P M G Ap ers On the applicabilit yof sc hema
in tegration tec hniques to database in terop eration in Pro ceedings Fif
teen th In ternational Conference on Conceptual Mo delling ER Cot
tbus German y Berlin SpringerV erlag The role of in tegrit y constrain ts in database in terop eration
in Pro ceedings nd In ternational Conference on V ery Large Databases
VLDB Bom ba y India San Mateo CA Morgan Kaufmann Pub
lishers pp U Da y al and HY Hw ang View denition and generalization for
database in tegration in a m ultidatabase system IEEE T rans Soft w are
Eng no pp No v

Linked assets

Computer Science Technical Report Archive

Conceptually similar

PDF

USC Computer Science Technical Reports, no. 636 (1996)

PDF

USC Computer Science Technical Reports, no. 639 (1996)

PDF

USC Computer Science Technical Reports, no. 899 (2008)

PDF

USC Computer Science Technical Reports, no. 966 (2016)

PDF

USC Computer Science Technical Reports, no. 632 (1996)

PDF

USC Computer Science Technical Reports, no. 810 (2003)

PDF

USC Computer Science Technical Reports, no. 969 (2016)

PDF

USC Computer Science Technical Reports, no. 763 (2002)

PDF

USC Computer Science Technical Reports, no. 669 (1998)

PDF

USC Computer Science Technical Reports, no. 641 (1996)

PDF

USC Computer Science Technical Reports, no. 943 (2014)

PDF

USC Computer Science Technical Reports, no. 749 (2001)

PDF

USC Computer Science Technical Reports, no. 851 (2005)

PDF

USC Computer Science Technical Reports, no. 753 (2002)

PDF

USC Computer Science Technical Reports, no. 633 (1996)

PDF

USC Computer Science Technical Reports, no. 638 (1996)

PDF

USC Computer Science Technical Reports, no. 642 (1996)

PDF

USC Computer Science Technical Reports, no. 801 (2003)

PDF

USC Computer Science Technical Reports, no. 631 (1996)

PDF

USC Computer Science Technical Reports, no. 626 (1996)

Description

Mark Vermeer. "Extending the remote exchange framework to suit instance-based." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 643 (1996).

Asset Metadata

Creator Vermeer, Mark (author)

Core Title USC Computer Science Technical Reports, no. 643 (1996)

Alternative Title Extending the remote exchange framework to suit instance-based (title)

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)

Tag OAI-PMH Harvest

Format 22 pages (extent), technical reports (aat)

Language English

Unique identifier UC16271175

Identifier 96-643 Extending the Remote Exchange framework to suit instance-based (filename)

Legacy Identifier usc-cstr-96-643

Format 22 pages (extent),technical reports (aat)

Rights Department of Computer Science (University of Southern California) and the author(s).

Internet Media Type application/pdf

Source 20180426-rozan-cstechreports-shoaf (batch), Computer Science Technical Report Archive (collection), University of Southern California. Department of Computer Science. Technical Reports (series)

Access Conditions The author(s) retain rights to their work according to U.S. copyright law. Electronic access is being provided by the USC Libraries, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Repository Email csdept@usc.edu

Inherited Values

Title Computer Science Technical Report Archive

Description Archive of computer science technical reports published by the USC Department of Computer Science from 1991 - 2017.

Coverage Temporal 1991/2017

Repository Email csdept@usc.edu

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)