' $ Logic Based Mo deling and Analysis of W orko ws A CM - - PowerPoint PPT Presentation
▶
SMART_READER_LITE
LIVE PREVIEW
' $ Logic Based Mo deling and Analysis of W orko ws A CM - - PowerPoint PPT Presentation
Oct 06, 2023 622 likes •837 views
' $ Logic Based Mo deling and Analysis of W orko ws A CM PODS'98 | | Hasan Da vulcu* Dept of CS Univ ersit y at Ston y Bro ok N.Y. 11794, U.S.A. * Join t with M. Kifer, C.R. Ramakrishnan, I.V.
SLIDE 1' & $ % Logic Based Mo deling and Analysis
f
W
rko
ws | A CM PODS'98 | Hasan Da vulcu* Dept
f
CS Univ ersit y at Ston y Bro
k
N.Y. 11794, U.S.A. * Join t with M. Kifer, C.R. Ramakrishnan, I.V. Ramakrishnan Hasan Da vulcu { Univ ersit y at Ston y Bro
k
1
SLIDE 2' & $ % W
rko
ws W
rko
w: A collection
f
in ter-related tasks and transactions designed to carry
ut
a w ell-dened business pro cess. W
rko
w Managemen t: Automated c
r
dination
f
w
rk,
among pro cessing en tities, to ac hiev e an
ver
al l business go al. W
rko
w Managemen t System (WfMS): System for automation
f
w
rko
w pro cesses (lik e DBMS facilitates creation and main tenance
f
large data sets). Hasan Da vulcu { Univ ersit y at Ston y Bro
k
2
SLIDE 3' & $ % What Kind
f
Computations are W
rko
ws ?
Long-running:
Hours, da ys, mon ths;
Autonomous,
distributed pro cessing en tities;
T
ransactional Seman tics. Hasan Da vulcu { Univ ersit y at Ston y Bro
Global Coordination Dependencies: Control Flow Graph:
❏
Hasan Da vulcu { Univ ersit y at Ston y Bro
k
4
SLIDE 5' & $ % Cen tral W
rko
w Problems
b c d a
XOR
1. d before c
2. b AND IF occurs (b) THEN c
Consistency: Is con trol
w
graph and co
rdination
dep endencies c
nsistent
? Correctness: Do es the sp ecication satisfy certain k ey prop erties ? e.g., Ev ery bid is either accepted
r
rejected b y the bid-ev aluation w
rko
w. Co
rdination:
Ho w to sc hedule tasks automatically ? Hasan Da vulcu { Univ ersit y at Ston y Bro
k
5
SLIDE 6' & $ % What is Needed to Enable Workow A utomation ? Creation: A formal language to sp ecify the structure
f
pro cesses and their in teractions at a high-lev el; V erication: R e asoning metho ds to ascertain that a w
rko
w is correct; Execution: T ec hniques for automatic al ly deriving correct executions from high-lev el sp ecications.
Ideally
, all should t in a single, unifying framew
rk.
Hasan Da vulcu { Univ ersit y at Ston y Bro
k
6
SLIDE 7' & $ % Requiremen ts for W
rko
w Sp ecication
Primitiv
es for mo dular sp ecication { Se quential and c
ncurr
ent comp
sition;
{ Subr
utines
(e.g. sub-w
rko
ws); { De clar ative queries (e.g. transition conditions); { A tomicity; { Isolation;
T
emp
r
al c
nstr
aints for expressing co
rdination
dep endencies.
T
riggers (e.g. exceptions) Hasan Da vulcu { Univ ersit y at Ston y Bro
k
7
SLIDE 8' & $ % Related Researc h
n
W
rko
ws
Extended
T ransaction Mo dels { P ermits hierarc hical structures and relaxes A CID prop erties. [Elmagarmid-90]
T
ask Dep endencies { Declarativ e constructs suitable for sp ecifying global co
rdination
dep endencies. [Klein-91, Singh-96]
Activ
e Rules and T riggers { Suitable for sp ecifying reactiv e b eha vior. [Da y al et.al.-91,96]
P
etri Nets and T emp
ral
Logics { Suitable for mo deling and v erifying c
ncurr
ent pr
c
esses. [Sheth et.al.-93, Nabil et.al.-98] Hasan Da vulcu { Univ ersit y at Ston y Bro
k
8
SLIDE 9' & $ % Our F ramew
rk:
Concurren t T ransaction Logic (C T R )
C
T R : Conserv ativ e extension
f
rst
rder
logic for programming , executing and reasoning with state-c hanging concurren t pro cesses.
C
T R uniformly mo dels: { Declarativ e queries { Concurren t up dates { T ransactional b eha vior { Mo del The
ry:
A
mo del assigns transaction form ulas truth values
v
er p aths. (i.e., nite sequences
f
states)
Informally
,
b
eing true
v
er path
means
\ can execute along
."
{ Pr
f
The
ry:
Sound
and complete.
Constructs
exe cution p aths as it pro v es statemen ts. Hasan Da vulcu { Univ ersit y at Ston y Bro
k
9
SLIDE 10' & $ % C T R : Logical Connectiv es
cost
up date
budget
up date means, \Execute cost update and then execute budg et update".
D D D D D D D D 1 2 3 4 5 6 7 8
cost_update budget_update
(con
tractor
nancial)
j (consultan t
billing
) means, \Execute (contr actor
f
inancial ) and (consul tant
bil
l ing ) in an interle ave d fashion".
D D D D D D D D 1 2 3 4 5 6 7 8 contractor consultant financial billing
( contractor financial ) ( consultant billing )
Hasan Da vulcu { Univ ersit y at Ston y Bro
k
10
SLIDE 11' & $ % C T R : Logical Connectiv es
bid
ev al ^ c b efore f means, \Execute bid ev al and c bef
r
e f along the same path". (i.e. ^ can b e used to express constrain ts
n
execution.)
D D D D D D D D 1 2 3 4 5 6 7 8
bid_eval c_before_f
in
ternal ev al _ external con tractor means, \Execute either inter nal ev al
r
exter nal contr actor ".
:
external con tractor means, \Do an ything but an execution
f
exter nal contr actor ".
db
up dates cost up date
budget
up date means, \An execution
f
cost update
budg
et update is also an execution
f
db updates". (
is
dened as
_
: ).
db
up dates means, \Execute db updates in isolation". Hasan Da vulcu { Univ ersit y at Ston y Bro
k
11
SLIDE 12' & $ % Putting It Altogether : Bid-Ev aluation W
rko
w
Con
trol-o w graphs translates straigh tforw ardly in to CTR form ulas: bid ev al r
(f
inancial j db updates j technical )
r
est f inancial
([o:ev
al = \hig h"]
f
) _ (l
w
f
) db updates (c
[c:cost
< budg et]
b)
s
technical (t
i)
_ (e
m)
_ (t
i
j e
m)
Global
Co
rdination
Dep endencies can b e mo deled as conjunction
f
a set
f
dep endencies C . 1: Ol
w
! :Oe 3: Ot ^ Oe ! Oe
Oi
2: Oe ! (Oo
Oe)
4: Oc
Of
( O means \
c
curs sometime".)
W
rko
w sp ecication: bid ev al ^ C Hasan Da vulcu { Univ ersit y at Ston y Bro
k
12
SLIDE 13' & $ % T emp
ral
Dep endencies 1. Existence constrain ts: Oe, :Oe (e m ust/m ust not
ccur);
2. Serial constrain ts: Oe
Of
(e m ust
ccur
b efore f ); 3. Complex constrain ts: C 1 ^ C 2 , and C 1 _ C 2 , if C 1 , C 2 are constrain ts..
Constrain
ts are C T R form ulas. Hasan Da vulcu { Univ ersit y at Ston y Bro
k
13
SLIDE 14' & $ % Apply T ransformation
Executing
a w
rko
w G ^ C b y the general pro
f
theory has exp
nential
run-time complexit y .
W
e dev elop a re-write system, Apply , whic h transforms G ^ C in to an equiv alen t ^-free form ula G C suc h that: { Exe cution and veric ation b ecomes more ecien t
n
G C . Hasan Da vulcu { Univ ersit y at Ston y Bro
k
14
SLIDE 15' & $ % W
rko
w Analysis with C T R Prop
sition
(Apply): C ^ G
Apply
(C ; G )
8
< : G C if satisable and G C is ^-free f al se if C ^ G is unsatisable
A
legal execution
f
C ^ G can b e pic k ed from G C in linear time b y the inference system. 2 { Inconsistency: I Apply (C , bid ev al )
f
al se { V erication: Giv en a prop ert y : Apply (:; G C )
8
< : f al se if
holds
for G C G C ^: all coun ter executions
f
Inc
nsistency
and veric ation reduce to the same logical problem { unsatisabilit y. Hasan Da vulcu { Univ ersit y at Ston y Bro
k
15
SLIDE 16' & $ % Apply T ransformation: Con td.
γ α η δ β
XOR
Example: If T is
(
_
_
)
,
then Apply (O ; T ) = O ^ T =
Apply
(:O ; T ) = :O ^ T =
(
_
)
Hasan
Da vulcu { Univ ersit y at Ston y Bro
k
16
SLIDE 17' & $ % Apply T ransformation: Con td.
ρ ρ
1 n
AND
γ β α δ
XOR XOR
Example: Apply ( O
O
; ( _
)
j ( _
)
j
1
j ::: j
n
) = (
send(
)) j (r eceiv e( )
)
j
1
j ::: j
n
Hasan Da vulcu { Univ ersit y at Ston y Bro
k
17
SLIDE 18' & $ % Knot Detection and Elimination Insertion
f
send/r e c eive ma y cause a cyclic w ait, whic h w e call a knot.
❏
1. I before O
2. If occurs (E) then O before E
3. If occurs (T) AND occurs(E) then E before I
R AND T I E O
2
3 1
Wf
OR
Receive Bid Evaluation Risk Consultant External
D
Decision Final Analysis Contractor Evaluation Technical
w sp ecication G ^ C is inconsisten t i Excise(Apply (C ; G ))
f
al se. Theorem (Corr e ctness) There is a constructiv e w a y
f
v erifying w
rko
w prop erties with Apply . Theorem (Co
r
dination) Let jG j denote the size
f
a con trol-o w graph, then w e can pic k a legal execution from Excise(Apply (C ; G )) in time line ar in jG j. Theorem (Complexity) Let jG j denote the size
f
a con trol-o w graph, N b e the n um b er
f
constrain ts in C , and d b e the largest n um b er
f
disjuncts in a constrain t.
The
w
rst-case
size
f
Apply (C ; G ) is O (d N
jG
j). (where as standard mo del-c hec king algorithms for this problem are exp
nen
tial in jG j)
F
r
certain classes
f
constrain ts Apply (C ; G ) is linear in jG j.
Excise(G
) is linear in jG j. Hasan Da vulcu { Univ ersit y at Ston y Bro
k
19
SLIDE 20' & $ % F uture W
rk
F ailure Managemen t: F acilitate failure detection and handling with adv anced features lik e con tingency and comp ensation; Exception Managemen t: Impro v e supp
rt
for triggers to facilitate r e active b eha vior due to exc eptions; Data V alue Dep endencies: Include transition conditions in analysis; W
