Intro Enterprise Comp
Intro Enterprise Comp CS 4365
Popular in Course
Popular in ComputerScienence
This 0 page Class Notes was uploaded by Alayna Veum on Monday November 2, 2015. The Class Notes belongs to CS 4365 at Georgia Institute of Technology - Main Campus taught by Staff in Fall. Since its upload, it has received 9 views. For similar materials see /class/234160/cs-4365-georgia-institute-of-technology-main-campus in ComputerScienence at Georgia Institute of Technology - Main Campus.
Reviews for Intro Enterprise Comp
Report this Material
What is Karma?
Karma is the currency of StudySoup.
Date Created: 11/02/15
Research Issues in Large Work ow Management Systems Gustavo Alonso Ilans Jorg Sehek Database Researeh Group Institute for Information Systems ETH Zentrum Zurieh CPI 8092 Switzerland Email alonsosehekinf ethz eh April 30 l 996 Abst raw 1 n this position paper we deseribe what we believe are I39umlamental weaknesses ol39existing eommereial work ow produets and bow database teebnology ean be used to address these issues By exporting database teebnology beyond the transaetional domain we expert to be able to design sealable robust I39ailure resilient and llexible work ow systems to be used in applieation domains ranging from busine f 7 proeess to eomputer integrated manul39aeturing 1 Introduction Workflow Management Systems are quiekly beeoming the teehnology of ehoiee to implement large and heterogeneous distributed exeeution environments where sets of interrelated tasks ean be earried out in an e ieient and elosely supervised fashion In many a workflow management system is not different from a sophistieated seheduler in whieh the seheduling is performed based on task dependeneies organizational strueture staff availability and existing eomputing infrastruetme It is preeisely this eharaeteristie that makes workflow management systems so appealing as it makes them mateh very well eurrent organizational and teehnologieal trends deeentralization of the eorporation deeentralization of the deeision making the need for very detailed information about every day aetivities and the inereasing availability of distributed proeessing teehnology WWW CORBA OLE Java Proof of the ever inereasing interest in workflow management is the large number of eommereial produets that have appeared in the last few years These produets are nding an eager market and opening up important researeh opportunities In partieular there are many instanees in whieh the expeetations from the users and the aetual features provided by the systems are not well eorrelated The main reason for this is that the requirements of a workflow system in terms of sealability and system wide reliability exeeed those of database and transaetion proeessing teehnology Gaw94 whieh is the eurrent state of the art As a eonsmuenee workflow teehnology is far from being as mature as existing transaetion proeessing and database teehnology and laeks the performaee fault toleranee and flexibility that is provided by today s databases The obvious ehallenge is to extend existing teehniques and develop new ones to make workflow management the baekbone of the information proeessing teehnology of the enterprise The not so obvious question is l whieh teehniques are useful in a worlltflow environment In this position paper we outline what we believe is the relevant teehnology that ean be used to enhanee the funetionality of worlltflow management systems We motivate our ideas by pointing out the weaknesses in existing eommereial produets and briefly deseribing how databases and transaetion monitor teehnology ean serve as the basis to address these problems 2 Limitations of Existing Systems In the worlltflow area and in spite of the many researeh projeets underway the state of the art has so far been determined by eommereial produets Henee to understand the state of the art it is important to understand where worlltflow eomes from The origins of eommereial worlltflow systems ean be traeed baek to o iee automation image proeessing or eomputer supported eooperative work Hol94 The emphasis in these environments has been muting sharing and cooperation Issues sueh as performanee sealability or reliability are hardly ever eonsidered by this type of systems KAGMQG and these eharaeteristies have been inherited by worlltflow produets There are no eommereial worlltflow produets based on OLTP On Line Transaetion I roeessing or database teehnology although many of them use databases as the underlying repository and some ineorporate ideas that ean be related to funetionality found in eommereial transaetion monitors As a eonsequenee the robustness and teehnologieal maturity reaehed in the database area is all but laellting in workflow systems Gaw94 Nowadays there several hundred eommereial produets that elaim to be worlltflow tools Of these only a handful are true worlltflow engines Some of the most relevant systems in the market inelude Action Work ow System of Aetion Teehnologies IBM s FlowMark WorFlo Business Systems of FileNet InConcert produeed by XSoft a division of Xerox Corp OmniDesk of Sigma Imaging Systems Ine ProcessIT of ATampT Global Information Solutions formerly NOR Sta ware of Staffware Corporation Regatta of Fujitsu eurrently ineorporated into ICL s TeamWA RE and OPEN work ow a WANG s produet It is also important to mention that there are a multitude of other produets being developed as third party applieations on top of distributed platforms sueh as LotiIsNotes Sueh produets play a role similar to the many third party tools used to interfaee with a database management system SQL forms for instanee and should not be eonsidered as worlltflow engines As has been pointed out many times in the literature GHS95 the rst generation of worlltflow engines is started to find wide aeeeptanee but available produets are far from providing optimal solutions The following are some of their most important limitations 0 Interoperability existing systems are almost totally ineompatible The situation is similar to that of databases before the widespread aeeeptanee of the relational model In spite of efforts like those of the Workflow Management Coalition Hol94 eurrent produets ineorporate in the design very eonerete and exelusive interpretations of the world AetionWorkflow being the best example mallting praeti eally impossible to federate different systems These ineompatibilities are not just the syntax or the platform but the very interpretation of worlltflow exeeution Some other systems are so tied to the un derlying support system that is unfeasible to extend their funetionality to aeeomodate other worlltflow interpretations for instaee worlltflow tools developed on top of Lotus Notes o Scalability as eooperative tools existing workflow engines have been designed with small groups in mind Only afterwards when users realized the potential of workflows these engines have been applied to large seale environments However the inherent restrietions in the designs prevent existing systems from being able to eope with a fraetion of the expeeted load whieh ean be as high as several hundred thousand proeesses and tens of thousands of users KAGMQG 0 Availability the degree of resilienee to failures of enrrent systems is almost nnll Cnrrent prodnets have a single point of failure the database and no meehanism for baeknp or e ieient reeovery This is not as mneh a flaw as a design deeision sinee these prodnets were initially intended for small groups and small loads 0 Correctness of exeention of several workflows has not been eonsidered an issue so far As long as we deal with loosely eonpled systems where no integrity eonstraints exist that span multiple systems the single steps of a workflow ean be exeented without any further eontrol If however in a more tightly eonpled system dependeneies must be observed nneontrolled interferenees of several parallel workflows eonld be prevented with similar teehniqnes to those used in database transactions 0 Architecture the most eommon workflow design nowadays is a single database on top of whieh a single workflow engine provides serviees to several elients These arehiteetnres qniekly beeome a bottleneek Some enhaneed arehiteetnres have been proposed BMI94 AKAJVQ L AAE95 but so far they still remain researeh ideas In general enrrent workflow prodnets eannot be used as the basis for an enterprise wide system The basie elements of today s information teehnology within a eorporation are databases and on line transaetion proeessing In many the state of the art of eommereial workflow systems is far behind that and yet they are being presented as tools to handle environments mneh more eomplex and demanding than eorporate databases and transaetion proeessing 3 Databases and Work ows Many of the weaknesses deseribed above are still to be addressed and in some eases even to be reeognized as problems GH395 KAGMQG However there is a solid base of knowledge that ean be used to solve many of these issues Most of it involves extending database teehnology to the new applieation area In some eases it is not a straightforward extension sinee workflows have greater demands than databases but work in this area will both bene t workflow managament systems and databases Gaw94 As a rst step we suggest that these extensions shonld target the weaknesses disenssed in the previous seetion o Flexibility withont advoeating a standard sneh as SQL workflow teehnology is still too far from reaehing sneh a point it is very important that workflow systems be designed independently of the eharaeteristies of the workflow speei eation langnage I etri nets state eharts transaetional dependen eies et eetera A workflow engine must be flexible enough to exeente any type of workflow relationship o regardless of the it is speei ed To aehieve this it is neeessary to develop a mueh better under standing of the exeeution model of workflows Current researeh is still too bound to transaetion models whieh is very limiting when de ning generie workflows AAEJFQGI Interoperability database federation sehema integration and data warehousing are all relevant topies in the database eommunity Exaetly the same problems will exist with workflow engines as different parts of an organization develop their own workflow environments whieh later will have to be integrated Workflow arehiteetures need to be designed with this in mind It does not su iee to have a eommon model like the one proposed by the Workflow M anagament Coalition Hol94 as the problem involves mueh more eomplieated issues In this area teehnology like COBRA or OLE may have a eonsiderable impaet MSKWQG Availability very large workflow management systems will involve several thousand users hundreds of thousands of eoneurrently running proeesses and several thousand sites distributed over wide area networks These are aetual requirements on eommereial systems KAGMQG In these environments eontinuous availability is erueial in the same that eontinuous availability is the key to many bank ing and eorporate database applieations Database baekup and replieation teehniques eould be used to automatieally transferred the load of a failed eomponent to other servers in the system without requir ing to stop ongoing exeeutions Moreover this also implies a built in meehanism for load balaneing redireeting requests to other servers when exeessive load may affeet the quality of serviee provided by a 80 Cl39 Concurrency Correetness of parallel transactions in databases is ensured by serializabililty This is a too narrow de nition of eorreetness not suitable for workflows environments beeause workflows span multiple systems and their exeeution may take several days or weeks For similar reasons the QPC protoeol standard for distributed transaetions and the basis for transaetion proeessing monitors may not be suitable for workflows If applied subsystems eould be bloeked for long periods of time Instead open nested transactions eould be used to eommit subtransaetions early and avoid bloeking due to low level eonfliets if the higher level operations do not eonfliet This approach separates the responsibilities The eorreetness of single steps is under the responsibility of the subsystem that exeeutes the step while the eorreetness of the whole workflow is delegated to the more abstraet workflow BDSHJS has proposed to eombine this teehnology with workflow speei eation In DSW94 SSW95 these ideas are level where the information is available what interleavings should not be allowed used as an alternative for the QPC protoeol in distributed transaetion proeessing Exception Handling In workflow environments the system eannot arbitrarily abort an operation If an exeeption arises human interaetion is often required to resolve it as it is generally unwise to expeet the workflow designer to be able to prediet all possible exeeptions in the exeeution of a proeess In many eases given the nature and eharaeteristies of the applieations these exeeptions are unpredietable and range from simple databasolike eonfliets to ehanges in personal ehanges in legislation and eeonomie adjustments This is a departure from standard database transaetion teehnology sinee the system does not have the knowledge to resolve all possible exeeptions The proper level of abstraetion to resolve 4 this problem is the workflow level where at least a notifieation meehanisms must be ineluded Similar issue arise in the quot 39 of 39 1 39 of CIM subsystems NSSW94 and some of the ideas developed in that are eould be applied in workflows Moreover the problem of having a reeoverable state for long 1 39 F l to 39 l l a flexible exeeption handling meehanism is also started to be addressed in the transaetion proeessing aresISTQGI o Faulttolerance in large seale workflow systems there will be a variety of eomponents involved in the exeeution of every aetivity within a proeess The system must tolerate failures of any of these eomponents For instanee when a workflow engine is not available it should be possible to automati eally eonnoet the elients to other engines to eontinue the exeeution of the proeess Depending on the type of failure the worst seenario should be that a proeess exeeution is interrupted until the failure is repaired lUpon reeovery exeeution should be resumed at the same point where it interrupted Some work has been done in this area BMRM AKAJVEM AAE95 but mueh remains to be done 0 Performance existing systems do not eonsider performanee to be an issue But it is possible that all or a great part of a workflow proeess is solely eomprised of automatie tasks with no human intervention in whieh ease the workflow engine beeomes the major souree of delays in the exeeution of a proeess While there is a trade off between the flexibility provided by workflow systems exeeution not based on eompiled eode and the e ieieney of eonventional systems like TP monitors where the exeeution is based on a eompiled program the system should be able to provide meehanisms to enhanee the performanee when automatie tasks are involved The need for high performanee distributed proe sing is not exelusive of transaetion monitors Gaw94 o Scalability the major drawback of existing systems is that they are not able to eope with large seale environments In the best possible seenarios eommereial systems ean be used with up to 40 users and a few hundred proeesses running eoneurrently This is far from the figures one eneounters as common requirements To taekle realistie applieations the sytem designed should be easily sealable to any number of users and eoneurrent proeesses This implies that replieation teehniques must be ineorporated within the arehiteeture and all eomponents must be designed in a modular fashion to allow eustomization of the system 4 Conclusions While databases guarantee the safe storage and easy aeess to massive amounts of data workflow manage ment systems are intended as the basie support for information flow in those same environments where databases are used However the eurrent state of the art of workflow engines is far behind that of databases and transaetion proeessing The sueeessful ineorporation of workflow systems into the eorporation greatly depends on the ability to make workflows a teehnology as mature and resilient as existing databases In this position paper we have deseribed what in our opinion are important weaknesses of eommereial produets Our approaeh to these problems is to extend database teehnology to the new applieation environment with out neeessarily seeing workflows as a database applieation Workflow management systems ean benefit from oxisting tochnology but also roquiro now solutions as thoir domanols aro considorablo largor than thoso of tod ay s databasos References AAE95 ME96 AIA94 180931 33an DSW94 Gaw94 GHSQ5 H0194 mama G Alonso D Agrawal A El Abbadi C Mohan R Gunthor and M Kamath ExoticaFMQM In PIP Wng Working Confrmwm on Information Systvm Davvlopmmt for Dmmtralisvd Organiza A I orsistont MossagoHasod Architocturo for Distributool Workflow Managomont tirms Trondhoim Norway August 1995 G Alonso D Agrawal A El Abbadi M Kamath H Gunthor and C Mohan Adyancod Transaction Modols in Workflow Contoxts In Pmcmdmgs of tin 122th Intvmatirmal Confrmwm on Data Eazgimvrmg Now Orloans Louisiana USA Fobruary 39l996 G Alonso M Kamath D Agrawal A El Abbadi H Gunthor and C Mohan Failuro Han olling in Largo Scalo Workflow M anagomont Systoms Hosoarch Hoport H3 9913 IBM Almaolon Hosoarch Contor Noyombor 39l994 Y Hroitbart A Doacon H I Schollt A Shoth and G Woikum Morging Application contric and Data contric Approachos to Support Transaction oriontod Multi systom Work ows ACM Sigmod Rmord Soptombor 39l993 D Barbara S Mohrota and M Husinkiowicz INCAS A Computation Modol for Dynamic Workflows in Autonomous Distributool Enyironmonts Tochnical roport M atsushita Information Tochnology Laboratory April 39l994 A Doacon H I Schollt and G Woikum Somantics basool Multiloyol Transaction M anagomont in Fodorayool Systoms In Pmcmdmgs of tin 10th Intvmatirmal Confrmuum of Data Engimvrmg Houston Toxas lUSA Fobruary 39l994 D Gawlick High I orformanco TP Monitors Do Wo Still Noool to Doyolop Thomrgt BEE Bullvtm of tin Thcha cal Committm on Data Engimvrmg HUNG 2 March 39l994 D Goorgalltopoulos M Hornicllt and A Shoth Procoss Modoling to Workflow Automation Infrastructuro Distributvd and Parallvl Databasvs 32 9 153 April 1995 An Oyoryiow of Workflow M anagomont From D Hollinsworth Tho workflow roforonco modol Tochnical Hoport TCOO IOOS Workflow M an agomont Coalition Docombor I994 Accossiblo yia http aiai ool ac ilkWiMC M Kamath G Alonso G Gunthor and C Mohan I royiding High Availability in Vory Largo Workl ow Managomont Systoms In In Procmdmgs of tin Fifth Intvmatirmal Confrmwm m Emtmdmg Databasv Tmhawlogy EDBT 96 Ayignon Franco March 39l996 Also ayailablo as IBM Hosoarch Hoport H39967 IHM Almadon Hosoarch Contor Iuly 39l995