Special Topics CS 8803
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 8803 at Georgia Institute of Technology - Main Campus taught by Calton Pu in Fall. Since its upload, it has received 97 views. For similar materials see /class/234064/cs-8803-georgia-institute-of-technology-main-campus in ComputerScienence at Georgia Institute of Technology - Main Campus.
Reviews for Special Topics
Report this Material
What is Karma?
Karma is the currency of StudySoup.
You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!
Date Created: 11/02/15
CS48038803 soon CS43656365 0 Enterprise Computing 12A Open urce Software Instructor Ca ton Pu unofficial TA Lenin ingaravelu Tradition terprise Software 0 Having someone to yell a hen something goes wrong 7 If a vendor is too small to solve a p b may choose to fold instead 7 How much liability insurance do you ha 7 How much are you willing to pay for it 0 Can we have free enterprise software Examp f Free Software 0 Unix operating systems 7 Free BSD Linux 0 Relational database management 7 MySQL Berkeley DB also XML 0 Internet and web software 7 ApacheTomcat web server 7 Boss app server JONAS etc o 1969 Unix by 8 Berkeley BSD ThompsonRitchie 39 o 1986 Posix standard 0 1989 SVR4 ATT and SUN 1987 Minix o 199X SUN Solaris 7 Andy Tannenb m o 1991 Linux 7 Linus Torvalds 7 1994 Linux 10 7 1996 Linux 20 7 2001 Linux 24 7 2003 Linux 26 e Development 0 Other complications in the process 7 Lack of centralized control and quality co trol 7 Questions of stability and evolutionary path 7 Tradeoffs between free software and high maintenance costs TCO Source Models 0 In analogy to scienti c re 7 Done with little monetary incen 7 Build on open standard interfaces 0 In combination with outsourcing 7 Distributed development 7 Build on welldefined interfaces and isolate components IBM a 0 Linux 7 The only OS on all IBM platfo o Alphaworks 7 Eclipse Java development platform 7 Many other examples e g tools for man ing WebSphere Credits 0 MySQL AB Zak Breant 0 Somewhat dated 2002 7 Optimized for speed reliabilit narrower deeper feature set 7 Flexible TransactionalNontransacti storage engines 0 Extensively used worldwide more than 10 000 server downloads per day 7 Low cost of ownership Duallicensed under GPL and Commercial licenses 7 Wellsupported and documented easy to install configure and manage o I 99 7 First commercial licenses agreement sold 0 2000 MySQL license changed to GPL GN U Public License 0 2001 MySQL corporate WySQL AB 7 Headquarters Uppsala Sweden 7 Lead investor ABNAmro big bank 0 Commercial Licences 0 Support Agreements 0 Partnership Agreements 0 Training Courses 0 Consulting 7 MyISAM has very compact tables Powerful features 7 Localized to many languages Japanese Chine German 7 Master Slave Replication for High Availability 7 Per table choice of backend withwithout transactio s 7 Embeddable engine no client server 7 Powerful security system o ADA o C 0 C 0 Common Lisp o Delphi 0 Dylan 0 Guile 0 And many more from the very beginning 7 CREATE TABLE TYPEHEAP 7 ALTER TABLE 7 Static dynamic and compresse adonly row formats but no transactions 7 Text and compressed indexes 7 Data and indexes in separate files 0 Especially useful for websites amp loggin 7 Fast readwrite performance but low rw concurrency 7 Good concurrency in the selchappend 7 External check and repair myisamchk 0 Full transactions A CID 7 With row level locking 7 Better concurrency than MyISAM on the same table 7 Consistent reads Oraclestyle M VC C 7 Uses tablespaces instead of individual le oMySQL AB provides support for InnoDB through a contract with the creatorsdevelopers o Is included in MySQL 4 amp the MySQL Max Stora ngine Merge 0 Merge tables is a collect1 of identical MyISAM tables that can be u table 0 Some of the tables can be compress o ALTER TABLE is used to change the et of tables 0 lnsertDeleteUpdate Select works 0 VERY useful for logging systems since MyISAM has perfect locking for this case 17 0 Original draw card 0 Completely inmemory with based indeXing 0 lnsertDeleteUpdate Select works 0 Useful for 7 Temporary tables 7 Lookup tables 0 Limited Specialized Storage ine BerkeleyDB 0 Full transactions A CID locking 39th page level 0 Better concurrency than MyIS readwrite on the same table 0 Primary key lookups can be faster tha MyISAM 0 Data and indexes in one le per table 0 Is included in the MySQL Max binary 0 Can also be used at a lower level independent of MySQL 0 MySQL supports many sites reliability 0 This is done by replicating the syste machines 0 Examples of users are 7 Yahoo Finance And other parts of Yahoo 7 Mobilede sells used cars in Germany Over 270 Million page Views per month with 49 MySQL slaV 7 slashdotorg o The single Master keeps bina log of SQL commands that update data a Slaves connect to the master or ano read and rerun the updates a In MySQL 40 the slaves use two threads 7 One to read the all queries 7 One to actually do the updates a This makes sure that every slave has all data e n if the master goes down while the slave is worki g on a slow query 7 6 web servers for dynamic con 7 6 web servers for static content 7 Each dynamic hit involves a database uery 7 Platform LAMP LinuxApacheMySQ HP 0 In June 2001 the site served 91M pages handled 342M hits max 1M hitshour 0 Hardware 15 Machines 7 Dual Pentium 600 Mhz with 1GB RAM 7 Quad Xeon 700 Mhz with 2 GB RAM My 4x Features 0 Reconstruction of le form management and table 7 Better scalabilty Faster and easie new features 0 Embedded server library 7 Embed MySQL in your application witho t a separate process or separate application SQL Table commands 7 UNION TRUNCATE DELETE and UPDATE w1 h multiple tables MS Access Syntax o More features for replication 7 Better safety and more features INDEX FIRST 7 Lowlevel interface for reading ta faster but simpler 0 Query cache 7 Improves performance in highrea situations Most websites has some q ry that will run many times a second 0 New fulltext search 0 Much faster bulk updates 0 Full Foreign keys 0 Change server options on the and globally 0 Secure connections with SSL 0 New Key cache with better concurrenc 0 Boolean operations for fulltext search 0 Online backup of MylSAM tables 0 Subselects o Failsafe replication MyO amp Windows 0 MySQL is actively deve Windows 7 But Windows is still not recomme e load platform UNIX performs much ette 0 MyODBC actively developed 7 Cursor support ed and tested on 7 Transaction support 7 Lots of small annoying long lived bugs fixed 0 ODBC 35 compliant driver released J anua Free 0 Independence 7 Access to all technical informa 39 a vendor chosen subset to become a expert as company employe s 0 Competing commercial services availab e 0 Lower investment in time and money 7 No extra fees for advanced features like replicationfree text search 0 Result Low Total Cost of Ownership 7 27 Free 0 Community 7 A 4439 39 testing etc resources 7 Find bugs faster on more platforms 7 Everyone wins original developers end use software projects 7 Trains and exposes skilled developers 0 Strong advantages for all users including proprietary users 7 Less risk 7 Deep integration is much easier CS48038803 ENC soon CS43656365 ntro Enterprise Computing Monitors and Reflective T Framework Instructor Ca ton Pu Unof cial TA Lenin ingaravelu a TP Monitor Message Manager Request Control Transaction Server g Display Transaction Server Page 1 0 System supporting transa between services 7 application servers process client requ ts and interact with resource managers 7 resource managers RMs manage recoverable application data 0 Missioncritical software 11 ds 7 Performance 7 Availability 7 Data Integrity 7 Security Authenticating users identity Authorizing requests for services Page 2 Client FrontEnds Display Replicated g Application Sewers Display Display Display Display Load balancing over replicated application server 4 Migration Display Display monitor Display server Centralized logging of information exceptions performa ce audit trail etc Dynamic Reconfiguration Page 3 W08 Problem Client 1 Client 2 I Client n Process In SERVER 5 Many Processes 0 Adverser affects OS ove 0 Too much processor context s 39 0 Consumes too much memory 7 May need paging IO 0 Distributionscalability add more proce ses 0 Hard to control load except by de activating clients Page 4 Wr Solution Client 1 RPC Process 1 Cl 2 Multithreaded lent Control Process Process 2 o Client n Process In SERVER 8 Advan s of TP Monitor 0 Multithreading gt Few p cesses 7 low OS overhead 7 less processor context switching 7 less memory overhead 0 Multithreading gt 7 easy to manage load by controlling m 0 RFC gt 7 easy to program distributed applications 0 Transactional RPC gt 7 easy to program distributed transactions Page 5 Dis ed Transactions 0 DB server only support mageneous distributed transactions ones that access ly that DB server product 0 Still need a TP monitor s transaction transaction can access 7 two or more DBMS or TP monitor products 7 recordoriented files 7 queues 0 The potential advantages of proprietary dist d transactions are performance and avoiding a T monitor for simple applications 0 Middle tier does 7 dynamic routing 7 parameterbased routing like partitio e DBs in a DB server 0 Reduces the number of clientserver ses 39ons 1 multiserver configurations 0 Supports queued requests Client Page 6 Applic Management 0 Partition applications indepe ent of the DBs they access 0 Prioritizing applications 0 Applicationbased load control and secu 39ty 0 Dynamic installation startup and shutdow of applications 0 Some TP monitors offer a lock manager and l manager for developing homegrown resource managers eg Transarc s Encina recovery unless there s a hot bac 0 Some DB servers don t have automat39 failover if a server fails 0 These automated recovery features are in ost TP monitors Page 7 o Transactional RPC RPC failure transaction abort 0 Log manager recovery data 0 Recovery manager log player 0 Lock manager concurrency control 0 Structured File System RM Page 8 En 0 Positive points 039 Evaluation 7 Logical component modularit 7 Transactional RPC 7 Nested transactions questionable val 7 DCE portability functionality 0 security portability multithreading RPC 7 Excellent callback mechanisms 0 Negatives 7 Performance problems on DCE Summ of TP Products 0 IBM family from 80 s to 7 CICS IMS 0 Open TP monitors during the 90 7 Encina Tuxedo TopEnd 0 Today s products 7 RDBMS Oracle DB2 SQL Server 7 App Servers WebSphere WebLogic Page 9 7 Wide range of extended transaction ode ETMs SpliUJoin ESR Sagas etc 0 Challenge 7 Numerous research papers hardly any realistic implementations CMohan SIGMOD94 Impleme ETMs is Hard 0 Complexity 7 OLTP facilities don t support 39 Join etc 7 Implementation is nonobvious o Practicality 7 Many legacy applications are happy wi ACID 7 Can t pick just one ETM o SplitJoin 7 longlived 0penended o Cooperative Groups 7 cooperative 7 Application domains are rapidly evolving Page 10 Reflective saction Framework 0 Practical 7 Systematic extension of OLTP 7 Concrete demonstration SpliUJoin ESR 0 Modular 7 Transaction Adapters as a thin layer 7 Build on available services to extent possibl 7 Concrete implementation on Encina TP Monitor Begin Splittid Jointid Commit Flat Spli Join Cooperative Grou Transactions Transactions Transactions 0 Unique set of control operations 0 Semantics associated with control operations Page 11 base interface Begin extended 139 Commit Abort CreateGro meta interface DelegateOp DelegateLock NoConflict TP Facility Similar39 39 among ETMs 0 Common extended functl 7 Relaxed con ict 7 Delegation 0 Split and Join transactions Nested transacti 7 Richer intertransaction dependencies o Cooperative Group Flex transactions etc 7 Structured relationships 0 MultiLevel transactions Cooperative Groups etc 0 Extended functionality layer over OLTP facility 23 Page 12 Architecture Transactmrml Appzzmzmm Transaction Manag ILock Managerl ILog Managerl ransactm Appzxmzmm c Splitl Join I CreateGrou Transaction Manager Adapter Transaction Manager Extended TP Fun anal y Con ict Lock Adapter Adapter Lock Manager Tran 0 Addon modules 7 Transaction manager adapter delegation of operations transactio relationships etc endencies 7 Lock adapter lock sharing lock delega 39on 7 Con ict adapter relaxed notions of con 39 t 7 Log adapter recovery information 0 Small set of adapterspeci c commands Transaction nagementAdapter Transaction Management Adapter TRACS 7 Transaction Adapter Command Set Encina TRAN module Maps commands in Transaction Adapter Comm d Set down onto TRAN operations and structures 27 Page 14 Transaction agement Adapter Transaction Management Adapter TRACS 7 Transaction Adapter Command Set Encina TRAN module Callbacks pass transaction information from T N up to Transaction Management Adapter structure 28 Transaction agement Adapter Ins tantiate create a descriptor f0 Uansac on Reflect extend the transaction Exec begin execution Delegateops delegate operations Formdependency form a transaction dependen y other commands Page 15 0 Dynamic transaction restructuri 7 release earlier modified data Split Ida T 1 T2 T1 T1 T2 Join m extended comman r splitjoin transactions EisplitOperationT2zTRID instantiateT2 reflectT2sj7model delegateilockT2 DelegateSet delegategppT2 DelegateSet T1 win execT2 return Transaction Split Operation I EiJoinOperationleTRID delegateilockTl DelegateSet delegategppTl DelegateSet commitself return Transaction Join Operation Page 16 3B Concurrency Co Instructor Calton Pu Unof cial TA Lenin Singaravelu 39 Transactions 0 Transaction Properties 7 Atomicity all or nothing 7 Consistency if updating DB prese e consistency 7 Isolation isolate effects of concurrent tr 7 Durability committed updates survive fail es 0 Sequence of data manipulation operation 7 Original ReadWrite 7 SQL Insert Delete Select Update Calton Pu Georgia Tech College of Computing Page 1 0 Transaction A o Withdraw 50 7 Read account 7 Result 80 7 Subtract 50 7 Subtract 20 7 Write 30 7 Write 60 0 If A writes last then 0 If B writes last the result is 30 result is 60 0 Bank lost 20 0 Bank lost 50 Con ency Control 0 First Law of CC 7 Concurrent execution should be sparent ie application programmers write e ential program 0 Performance requirements 7 Low overhead per transaction 7 High concurrency beyond serial execution 0 Correctness criterion 7 Must make sense 7 Formal definition required today Calton Pu Georgia Tech College of Computing Page 2 0 Sequential Execution 7 Satisfies ACID properties 7 OK for uniprocessor but limits sca b1 39 V 0 Concurrent Execution 7 InterleaVing actions overlap lost update 7 Correctness serializability SR 7 SR execution that produces result equivalen to a serial execution 7 Intuition of SR nondeterminism 7 CC algorithms that maintain SR 0 Transaction Model 7 Execution history of RW actio 7 Serializable histories 0 Transaction Dependencies 7 Input action read 7 Output action write 7 Action dependencies Calton Pu Georgia Tech College of Computing Page 3 e Description 0 Dependency Graph 7 Data ow through the actions 7 Collapse the actions into a transaction 7 Transaction dependency graph R80 60 W30 0 Cycle ltgt nonSR 7 Acyclic dependency ltgt partial or 7 partial order ltgt total order 7 total order ltgt SR 0 Kinds of Cycles 7 Lost updates see above 7 Dirty read read uncommitted updates 7 Unrepeatable read overwrite unprotected data 8 Calton Pu Georgia Tech College of Computing Page 4 o Locks 7 Read Shared WriteExclusive 7 Lock compatibility matrix 0 Well formed locks 7 Obtain appropriate locks before access 7 Unlock after use 7 Growing phase acquire locks 7 Shrinking phase release locks 0 Lock point locks needed by the transaction 7 Gives the serialization order 7 After growing and before shrinking hold 11 7 Separates Growing phase from Shrinking ph se Calton Pu Georgia Tech College of Computing Page 5 7 Sequence of lock point timestamps serial equivalent execution 0 Idea of the proof 7 Holding all locks force any dependent transactions to wait for this one to release lo they follow 7 2PL also forces preVious lock holders to precede Manager 0 Lock Manager interface 7 Lock names 7 Unlock names 0 Structure of Lock Manager 7 Multiple threads 7 Protected data structures 7 Encapsulated interfaces Calton Pu Georgia Tech College of Computing Page 6 7 Does not release locks increme 7 Release all locks at the end of trans t1 0 Tradeoffs compared to general 2PL gtLess concurrency hold locks longer ZEasier to implement ZNot a problem at low concurrency levels ZEasier to recover avoids cascaded aborts 0 Lock is free 7 Insert into the lock hash table e 7 Link into the transaction table 0 Lock is not free 7 If compatible shared similar to above 7 If not compatible wait with care 0 Unlock 7 Delete this request from both tables 7 Check change grant locks or not 7 Wake up new requesters if granting Calton Pu Georgia Tech College of Computing Page 7 committed data 7 In the real world relaxed isolation is adequate 7 Most of the time large part of DB always 1 committed state 7 Locking is useful would be useful to devise cheaper alternatives of Isolation 0 Degree 0 atomic writes 7 Wellformed wrt writes 0 Degree 1 No lost updates 7 Wellformed wrt writes 7 Twophased wrt exclusive locks 0 Degree 2 no lost updates or dirty reads 7 Wellformed wrt writes and reads 7 Twophased wrt exclusive locks 0 Degree 3 ACID transactions Calton Pu Georgia Tech College of Computing Page 8 0 Most practical 7 Almost all commercial DBMS 0 Performance advantages 2Relatively low overhead in CPU ZGood concurrency for low contention ZWellunderstood optimizations 0 Problems gt Deadlock handling gtDegradation at high data contention 7 Basic timestamps are assigned of each transaction 7 That ordering is maintained as the sen l or 0 Each data access is checked 7 Past accesses are recorded reads are updat 7 New access is allowed only if the timestamp ordering is legal Calton Pu Georgia Tech College of Computing Page 9 Overview of IXP2400 Architecture Intel 2 generation network processor 0 designed for network processing 7 goal to optimize for fast path and accommodate Slaw path 7 programmable 7 fast deployment of new services upgrades protocols I needs external control processor IXP2400 chip highlights 0 8 microEngiries uE 7 RISC 600MHz processors with instruction sets targeting networkingrelated operations 7 for fast path 0 XScale core 7 standard embedded CPU ARM family running Linux 7 slow path Media and Switch Fabric MSF Interface 7 to RxTx packets from physical layer andor switch fabric ie main data path interface 7 sepamte 32b Rx and Tx busses up to 125MHx w SkB buffers various con gs up to 4prs ours 3X1GbEthemet 7 connects to external devices and is con gurable for different bus protocols DRAM controller 7 for packet storage up to 2GB but access slow 7300 cycles 64bit wide SRAM controller 7 store packet queues lookup tables 7150 cycles 32bit Scratchpad memory 1 6k 7 on chip for parameters communication bw uEs PCI bus 64bit66MHZ 7 connect to host 7 external control processor or other PC1 compliant devices eg chain lXPs together Other units 7 Hash control registers other periphemls timers interrupt controller perf monitors John Morgan presentmran m A Microengine MEVZ 0 control store 7 for instructions 4k 40bit instruction uE 8 hardware supported threads 7 hide memory access 7 separate register sets PC 7 fast ctx switch 0 registers 7 256 GPR 32mm 512 Xfer 4x16thd7 DlSxRlW 128 NN 16thd 7 addressing context relative or absolute GPRs 7 local memory 640x32bits CAM 7 16 entry cache IPC messaging Via circular queues rings 7 scratchSRAM operations for ring management 0 signaling 7 15 signals per context 7 context can poll a signal explicitly 7 use to guard critical sections I o Other Local cans MicroEngine V2 Fk payluad n u CS48038803 ENC soon CS43656365 ro Enterprise Computing 39nual Queries Instructor Ca ton Pu unofficial TA Lenin ingaravelu 0 Continual Queries Projec 7 CQ concept implementation a 7 WebCQ demo 0 XWrap Projects 7 XWrapElite XWrapComposer 7 Omini object recognizer tivation B2B apps supply Chain mg n Event based information flow BZC and CZC apps Personalized fresh Push has to be intelligent e Detecting and noti ing ehanges Underlying technologies e utornatie detection and adaptation to changes wrapper component i Wrappers wrapper generators generators Example CQ 2 Trigger STOP Noti Trans Infe Q mative transport routes wind or rain Contin u at Qu ery Query Concept 0 Continual Query as quadr Q Trigger STOP Notificatio O Continual Semantics CQ issued once and run until STOP When Trigger becomes true Q is evaluat d New results of Q since the previous execu 39on will be returned The result is sent via personalized Notificatio CQ em Overview Internet Monitorin Filterin Notification by CQ CQ Engines Significant Results e Wireless 5 interval 0 Contentbased Triggers 7 When IBM stock is up by 5 the same da 7 Implementation feature Systemchosen pol 39ng interval sues in Web DB 0 Architecture for interoper e and scalable global information systems 7 Client and Server or PeertoPeer 7 mediatorwrapper or multiagent archr cture 0 Distributed query research issues 7 Query routing and catalog management 7 Distributed query optimization multi layere indexing query result assembly Researc esults Overview 0 OpenCQ 7 Update monitoring for structur structured data sources 7 Simulation Results 0 WebCQ 7 Update monitoring for arbitrary Web page 7 Demo Op m CQ httpdislccgatecheduCQ OpenCQ Architecture Core gt y 2 External LiuPuTangTKDE revised g ms sam munum o Trigger grouping Effectiveness on different types Effectiveness with respect to size 0 monitored How do different group size distribution affect the effectiveness of the solution 0 Multilevel CQ processing 0 Grouping cost analysis Expe ntal Platform 0 Hardware specs 7 Sun E450 server 4 400IVIHZ u 1GB RAM IOOMbps Ethernet 0 Runtime environment 7 Host OS Solaris 7 7 DB server Oracle9i 901 enterprise DB S 7 Java java 2 runtime environment java hotspottrn client VM build 130 Groupl ifferentTriggers Type 1 Trigger range test WHEN stockJast gt 58 WHERE stocksymbol Mspr 3m Yype2mggevYFtEIEI7 Emupmgcasks Nammupmg Emupaizesm a Emupaizess Type 2 Trigger radvhg up WHEN stockJast INCBYP 5 WHERE stocksymbol MSFT vamahm m swarms mqqzv Evaluatim m swarms Type 3 Trigger mutr39ge 5D rand122115 W 2 u 2mm mun sum mu muumuum DU 2 u 2mm mun sum mu munmuum nu wHEN stockjast gt 53 Numbevafcu NumbevaVEu AND smckvolume gt 20 000 000 r r I earn evltrFmu7Emu m mp5s r um evltrFmu7Emu m cast5E WHERE stocksymbol 7 MSFT sun M y W p g Nammupmg Emupaizesm 1 a Emupaizess Type 4 Trigger obi up WHEN Stock last CHANGES mqqzv mumquot m swarms Stocksymbol mun suuu anuu muumuuum an 2 u mm mm suuu anuu munuizuuum nu Numbev ar Eu Numbev a1 Eu Trigger 1 range test WHEN stockJast gt 58 WHERE stocksymbol MSFF Trigger 2 Wit7 aggregation WHEN AVGstockvoume gt 20000000 WHERE stocksymbol MSFF Grouping we mgw Ncqwuuu erent Source Sizes mggu W aggvegatmn Ncqsw Aouuu Nanremupmg Amt Nanrgmupmg 4r mupmw Hammer gm gm Sm gm gm W gm 2 u 500 mama m 3 u 2 u 500 mama m 3 u oamaum we m Kbym we mm Newmanme ck Emupmw a mvnuthui means32C MK oamaum we m Kbym mggev W waywaan Newmanme Emupmg EMU 2 u 500 mumuu man u 2 oamaum we m Kbym u snumuumuu 2mm 3 u oamaum we m Kbym MultiIe Trigger grouping Query caching Noti cation graupin CQ processing Mum mew gimpmg mm ms lbmhnun quot993 gmupslze39 0509 66 88 0 Human 1m Is cundsl B u x w davhmbn anmggm gmw SE9 n 0 9 Liwa 0 m Nun gmuphg 9 TG mags mepwgonm mac Hm 3cm ma CM cv lt G oup ng sha egy Mnqmupwg a m Ynggai Gmwmgn m 1 mac m1 30 3 I 160 um um Temcmn 40039va Sikh ebCQ WebCQ http disl cc gatech eduWebCQ We Architecture mm mbpagu Chung Summry Gznzrnliml Chung Nm39ru mn mm mmm 7 Structure Digest efficient enco 39 7 StructureAware Difference Algori RoccoButtlerLiu ZOOZ 0 Scalable Processing of Web Page Sen 39 7 User Profile Grouping of Web Page Sentin s with Structure Digest 39ng of Various Page 39 cts 0 Whole Page any change 0 Page components 7 Link 7 Image 7 Keywords 7 List 7 Table 7 Phrase 7 Regular Expression 0 Screenshots walkthrough 0 Live demo for WebCQ sentin 0 Wireless WebCQ 7 WebCQ for Palm 7 WebCQ for Wireless phones 0 WebCQ Extended Applications 9 Mom Fynnles ms ab slum em 2 w a Q HAD waw cc gale2h eelmneclsesvmw as Wequot Sentinel Ime Te Sand 39V 39A home CHHCCI39 E a e L credible Current Cumprehensive can555 A R led Prnjects m span el Search Results Yuur search retrieved 107 clinical trials fur Type af cancer nrnstate cancer State an states 1 e mi tna all t es cnuntry all cnuntraes npennctnve NIH Chm2 Center anly nn Vexslan af results nealtn nrnsessannal Vlmpley checked documents Q Title uf Clinical Trial 551332 pnase I Kandamlzed Stud af Gemsteln fax Preventlan af cancer in Patients Wlth a llcleaaecenuaa lnstnr af Cancer a Wlth Asymptamzbt Earl prestate cancer m m w 2 Pnase I Kandamlzed Stud at aenistein in Patients witln Sta e III ar IV Prastate Cancer uncelccceem llCIeGEIEIeHBB WCchKcelzna 3 pnase I Study af I rDlmenslana Canfaxmz Kadlathexagz in patients Wlth Lacallx Mchcealnzz Advanced Sta e 12 and 13 Adenacarclnama at the Prastate Summar Last Madlfled NCI VBI EIIEIZ mmnnm 9 nlemet He at Vew ngnlllgs Lungs J u Sen Address E nun waw cc gale2h edupmeclsmslcw Y W MW WW nae nun sentinel m Eai l Addre39 WW luti catrun letlnul Enail v latisicatian Frequency Muf n or r Emyr nays v sun canteen hurl Hum n We neeennen my Fjv v gm yr 39 Related Prnjects V a 3 hznf Ca CCfeL Credible Current Ccmprehensive Squot didr39an a nr n m muml mun ream Search Results Yuur search retrieved 107 clinical trials fur prastate cancer State all states all t es Cauntry all cnuntxles alienactive NIH clinical Center anly Vexslan at results nealtn nrnsessannai I lxplay checked documents protocol 1 Title uf Clinical Trial mms pnase I Kandamlzed Stud nf Gemsteln fax Preventlan af cancer 111 Patients Wlth a incrementgala Hlstax af cancer a Wlth As ptamatlc Earl Pmstate cancer Hw u 2 Pnase I Kandamlzed Stud at aenistein in Patients witln Sta e III ar IV Prastate Cancer uncelccceam V llCPGEIEFHBB Uquot totem glnlemet WE gm w ngn39anS Lars elu 55 j gd 9 mp 11W ee new edupmleclsdlslE M Getting client infurmatiun client HOST m2 gammedel 00 gatecll m 1311217 97 122 User authenticatiun setting User Emil ifsquotgiragafecll m Emil valiantiun paired Rt Parsing Sentinel Object etting Sentinel nane Prostate cancer 01111105 mel setting Sentinel Inxgel mu 1m cancernefncl1uh ivc minnone iiexevs ml Getting Sentinel lyue anycllange Gethng Sentinel Cuntent Banner Exchange uutaet c Search Rerulfsl Parsing Ilutificatiun Object mu Summary Kecagm ze lntnfncntnnn lethnd small 1quot nmm39b l xeeagnize lutisisutiun cunaitiun 1mm Parsing Step object hIHI39KLM mummn Kecagnlze smP comnlol 2111171277 12221 quot quotquot 1quotquot snulmu rumnui Installing sentinel un server mmquot undlllu c mechng tu WehCQ server Creatlng39z new Web sentinel With 10 9331321110 we Ma r Ilidon Creatlnz cache far the sentinel Updating WehCQ mnive list Activating sentinel sum cmmm squot umuuun Sentinel was installed successfully Sulu39e New tracking result summary rnim WehCQ11032000 z at November 3 2000 9 55 21 PM EST Fnun L L aLceh T g edu vagggcc gatech edugt W ilrir all l gle gel w v a Help 4 Wales a we Mupllel cc gale2h edu asawehcumuleslllmglmcc gab2h ecumama lm W bCQ c ptured new update an llnvelnber 5 2000 054 Pl PST L S nlinel 144821194Zriggr Evaluaiiun 2 Press Reland buttnll m ynux hxnwsuz 1f ynu saw llns page berm Senunel Type mychanze 1mm nu mt nmcunlnmchm nchc 1uXE7F nae Search Results Sentinel Content Cancer N e l39 A mre m H mm cmquot mumm Creamele Current Camprehensive J Cancer credr39lzle Current Comprehensive c er w New Emu4 mum Search Results Search Results Ynllr search retrieved 66 clillina trials fur Ynllr S l vh mu39iuvud 61 clinical win Inr Typa nf Emu pmmu cmwer NIH Chmcal Cantu unly lln Typu a meal Jinnah cmnr NIH Chnwal ocular uldy nn Type emu mzhnzm Versmnenesuus hzddlymfzssinml Typzafmal me Versmnenesuus humpmfzssipml smug arm npulhn vn Stags canon 11 sinng smug nflml analactive Slugs er s 39u Ill singly PmtacaHD Mommy ll PmtacaHD Mammy n my Drug lly mg Sm allstztzs ems afma z phzses Sm Illsth Phase mm memes 7an all caumles Spenser enml NCLspnnsnnd Caunuy ll ccmles Spnnsm emu NCLsyunsnmd Dlsplny checked document osplcy checked documenr Ilue nl Clininal Trial P g gt Illle nl Clinical Trial Pquot quot 1 l ths I Randmmzad Stud ultimatum fur Prevbn nn nf NCIQLCUUBBE 1 39 Thus I Rmdnm ad Stud nfOemsmn m Plovanunn a NCLBLCUUWE Cancexlnhuents WthNa Hlstag afCancex aleLh CancennPauents WlthNa Hlstag afCancex aleth l mm at Plnnu ax mO Mall 2 an 3940 End 1 at Cum moth Y 11 c 2 Phase l Randamlzedstud nfGemste gunPauents WLh ullclccc 2 r Phase l Rmaameasma arcemslemmmems wlm ullclccce Mlnlemel l Ens gall Mew genres pals g p asemll Agmess 1 HAD Mupllel cc gale2h edu asasMencumesulwlanecc gale2h ecumollan ml 3 211 nzwuplmcu r scmlacezclllvllmggcr zvaluz39 um Press Relmllbmcn m wmbmwsex ll39wn saw lls pag befnxe VA Sensingle myc mammal Hwea 39lzx cccmrmeclslAllml GA Us r 139 I 57 n m 51 A mg n n AypalTemp w 93 llqu 75 mm meh 39 13 Wm NW9 mph 1 l 39 vlem Iy I mi 5 w my W Dzwpuim 61 um um um zuusmes cw Ith new Emmtu nu sme new Ith cw mum seem nus mum A sun mm m mu NW mu 4 re than e m m Km mu mm ms m Whamcmg y ms mm 6751 mmmmmmem We all meme memenwmmn m l Nextyeav svveathevmwemhevglannevcam l m AlvOuaannd M Nextyeav swemhevmwemhevglannevcam l m AlrouamxanthgMAm Manszndl lies Mapsandlmzuas e c 7 27 39 r f 17 IR E M 439 nuclear mum ms use s m s We em mm me Yzhnu nesnurces Lucal rnrensmn39m me mulml Weather Semee Yzhnn Resnurces anzl Fnreczsl mm lhe mum Wemher Service Storm Center nulls canal ccemkee clam cube cewela d2 kale en Storm Center updated buns canal ccemkee clawsquot cube cewela dz WellherMEs gwmnen halalsunmamhemmaspennewlunmunhm WenherMES rmemcwmnemaralsun eavdmenNJaspevmeMuml W reckealesemmmlenspalmnemcluemcmecmesmalle W reckealeseummlenspalmnemcluememecmesmall W ccvmglcmccuclasvmecnmmawrencevmemanenamev lam cuvmgmnduuglasvmegumquotlawvenceleEmane amev Skl mm Skl mm mm muel on me mm EDTWED 00MB mun 39 quot 39 39 1nmPMEDIWEDOCI1EZ n New my Search New calSearch Tudav pamvsunnv nghsln heuppe s Nenrweslw memmrnmnswclezrmwsmme mm 5nsLiuhl nnrl swfh Tumgm clear Luws mu Nunhwmd lul mph seam n n e g ems use emcee u s c 95m mm emcee u swarm Thuvsday mus vsunnv Hmhslnlhemldms qumnum SmmevwrCA Thllrsnmnmht murmumlu see v d d lquot I 1 l l M Wee Ed at yew ngolle look new Agnew m1 JEo mu m captured new update on la 3 2000 80641 Al EDT lebCQ nlSllllZHtrigser evaluation quot1 Press Reload buuon in your brwser if you mine Tyne regexp quotIIDb Iovie of Yhe Day saw rhis vase before om eo m A mp 3992 m h 2quot A ewe w A gamma m r9 g valmus E Fa us1 anmmgngal gn rquot 1 IMDb Movle my Day 1 2110 n 5qu Ennllsh ml was 539 kindred sensibmzy mum e A Dmercnt Klnd o Credlt 1 4 r 7 alumna j 1 gee v Fewemer leue Ltng em Werequot an we mewzer ee gala2h eee seemseenIcureeeuwzeneeee galsch eee1721e15211a1 m1 2 men WehCQ czplund 112w update an 0mm 172nnn 7 2m PM EDT Seminal 172161524 1niggnr uvduminn 2115 Pm meme 1n ynmhmweux 11 ym1 law 1w pugn berm Seminal Tyye 111 Texan URL 1mg szadnzx me mmnlmcnmleATL 1m 1 u 0 a 1 1 Semie1 Curwere Cumm W931th Cand wns 011 11101 NEWPAGE Currem Weather Conditlons Currem Weather Conditlons r ran1 ran1 Slates mes 1111111414111 nmmzwzsm 1mm 33114111 nmmzwzsm eenemeus m 39quotww39 o 395 M EDT Cnlllllllnns at quotm 13 753 AM EDT zu u use u zuuu1 UTC vjnm 051m wmu vremme NNWraau eaereesy era MPH 5 Kr mewme a mums nsrhmw 1 me e Enn ns parl v mmv 5 send s penweeey Inmlmmlum 55 a F 1 a 0 Weather M er new anl 51 1 F we 0 Temperature 57 5 H11 1 c Relmwe 31 new Pnlmi n F 12 a c quotquotmm Relmwe ms pressure 3m 1n He 1mg We Humid emmeler Pmslum an ua 1n H1 11115 uh uh KATL171U532nnnnnKTesM rnguu some 1emme1en 13 m RM U OY EWEBTWGWE Pressure 1 n3 worms 1 a We Nehenhanmvea name see 1eneen nh KATL 1 e1 1 531 33mm 15M BR FEWZUD some 1013 Aauua RMKAO SLP1 95 T111 1111128 1111 57 211131 5111115 gdm Mew mm Inali p gum j39Agm s Q My Mupuev cl gale2h edu BSBSsew elMehE UEemev zsa 1 ma Installed Sentinels by user uangeccgmechedu Intn sentmels 2 Send me the sentme summary by small mun c Sexme um zuuuqrm nf ni Qneg Intlfxcatinn human 1 nunu mm aimed ma m y 1m iack mu an Yha 1 Madras hug fname1ahuacum975hmAdJIl we 3 mm mm samemum hum1W u gaucheauMmgczmw may Yyer d 5mg mychnge 4 m My Magma m h a m mum mm h 7 mm mum mm1 2nnn7m713 mm m mac mu 2nnnrmrn 1m fnanle mmw 751mm mg nan713u m pmme mm mm mummvsm 5 w h madam mm n v n N W31 ham yge znnnrmrn httEwmnc1mhgav ark mg u an m m hum4W nu mh gavgahumnmmaexmm we 2nnnrmrm 1 24 u Mm Def 10 sum 12 m1 ZDDEFEIFH httEwsathernaaagavweathexcunent AILhtm Me n 33u g gh emet htlpdisl cc gatech edu Web Re est for registration installation of sentinel or update Mobile Client Content adapted Web or cell phones Sentinel Installation c n Stock M Choose tar source URL Get updates Client Login CQ Applications 0 News stock ticker 0 Traf c monitoring 0 Web site aggregation 0 Interest Recommendation 7 We need to be able to access th 7 Access requires wrappers o Wrapper generation and maintenanc should be automated as much as possi 7 XWRAP Elite wrapper generator 7 Omini automated object extractor The cheapest price of such as XML relation table ltpncegtll 7Ultpncegt Pro Statement 0 Problem handcoding wr 7 timeconsuming 7 errorprone 7 hard to maintain 0 Solution wrapper generation ll er Construction amazoncoukw swans war max anawsz 5mm MEGMIES 53955quot PAPERMKS Quicksuncn Banks a mmumrs a ImERMEI Norwegian Wood Haruki Murakami Jay Rubin Translator 57 99 ltb00kgt 539 Yuu Sav 21 5D 2u Used PHEE 3 75 Functional Wr pers ltb00kgt Availability UsuaHy dispatched Within 24 hams Wm Calegury es F e ierder thm Ermeii a Friend Abnut This item ltOur7Pricegt1 00lt Ouripricegt ltb00kgt Wra r Generator 5mm ommtm 5 a mama editor anu The Wrapper Generator System I Kim i Slmrlurl Comer ram Analyzer irmiig mamiii 1 quotsummit umpptr r Proglilm Hankupiug39 WW4quot quot nauiim m 502mm Error Hnmllrr u A dln 39 VLk iHl x v wi mit Ulblmsrd Que Amer oi riimrr Comaquot nun in XML rappel rmgmrr Data er Generation Expert GUI Interface W Human Knowledge Extraction Rules Extraction Rules t mam mung Imd kanglug mvmppumam Wm mm mm Wmm c wwwmw 5mm 0 Region extractio Paragraphs Tables 0 Semantic tokens extraction Tokens of interest 0 Hierarchy extrac Nesting structure of sections ation Extraction anguisannm f1 mquot or 0m 5 HD 2 Boom WhatAmcntas mam Emmi u um Wm murm mmm v See Dem Related Informatiun Bibliography 1 Hamki Muvakami Hunks by Jay Rubin tion About this Item Descngtmn mm The Readers Cata ug me the Pubhshev ERRHON RE ISION HIAD Al HHtE Al HHGE THE IBEX CODHVG mml w RI KY SIZE THE MRI SIZE B 17133 MEX NOAA 40 min 2 7 days 7593 297 CIA Faa book 25 min 1 6 days 18981 260 Buy com 16 min 0 1 days 5172 148 Stoclamster 23 min 1 5 days 370 136 0 Pros 7 Semiautomated 7 Slow for at 39 7 Any Web Page pages 7 Handle complex data Require doma speci c knowle 7 High maintenance ost for certain format changes Document Output Object Separation Object Pruning Element Alignment O Automated Process Human Input 48 t Pruning 10 objects have 10 elements 8 objects have 12 elements 6 objects have 8 elements 3 objects have 1 element 2 objects have 20 elements lobject has 9elements lobject has 3 elements t Pruning 10 objects have 10 elements 8 objects have 12 elements 6 objects have 8 elements 3 objects have 1 element X 2 objects have 20 elements X lobject has 9elements lobject has 3 elements X 12 Elements Elements t Tagging Object 15 39 ct 16 15 16 httplink The ElephEstK nishes Stories http img Harukimu Dance Dance Dance A Novel Hardcover Haruki Murakami Elmer Luke Knopf Alfred A Hardcover February 1993 Kodansha America Inc January 1994 Ele tAIIgnment Object 15 ohiect 16 15 16 http k http img Dance Dance Dance A Novel The Elephant Vankkhes 1 ories Haruki Murakami Elmer Luke Harukimu Hardcover Hardcover Kodansha America Inc Knopf Alfred A January 1994 February 1993 Wrappe eneration Time 0 Evaluation of Wrapper Gener 39 Web sites Time mims esl Gold s Auction 10 Auctionscom 10 Cyberbuyer 15 Net Library 05 Egg Head 15 Foodgeeks 10 Bigfoot 15 Etoys Search 20 DBWorld CFP 10 7 Understanding the data structur 7 Looking for representative sample 7 Element Alignment Overhead Elite Wrapper Execution Time Time 5 Network Object Element Output Total Tlme Extraction Extraction Tagging Execmionsteps 0 Object Extraction 7 Success rate up to 93 in Omi 0 Element Extraction 7 Object Separation 90 to 100 for 7 Object Pruning 7 Element Alignment evaluating 0 Rapid aging as web pages evolve Accuracy Object Pruning Performnce Auctmns cum Freclslon seep 1 Step 2 Step 3 Step 4 1 Hm mu 5 Gene ID CMBWquot L mm 7 K nnem 0 Mn V ll WWW anAImy Analysis Clusl MmAnaly s ulm39m39z39 SEE 7 Ste 6 P P Step 5 GM mm m 7m WWW 1 lt m 741 WWW jm lm bmllmg hm Pmmumr Model gcncnnor anlmci Idemmcunun mm mme WW mum l Structural format such as XML relation table lttutalgt191lttotalgt quotquotquotquotquotquotquotquotquotquotquotquotquot quot Omini 0 Static web 7 Crawlable primarily as text 0 Dynamic web regular but not craw ble 7 Applications want to search dynamic we 0 7 Rapid changes not trivially detectable 7 Manually written wrappers mostly infeasible Omini ch Architecture w n W x w Dmmm W m w W WM 7 domain 75311er Web Sm qwrv um wk Imummml 39 mam mum Dx W 7 Automated analysis of the tree 7 Heuristics re ect commonality in 0 Content Region Identi cation 7 Find subtree with significant data objec 0 Content Object Identi cation 7 Find objects by significant separator tag mm onllm ma magma n 4 mm m Immu pllulrypmuuu mpmu a nquot mmn znm msm 4mm McuunurlnnHlnmu mun mam wms W nterest o Heuristics Highest Fanout HF W Content Greatest Size Largest Tag Count 4 h W lt7 mm A unnmmu m lm 5m mmmsarmzwmmmw mm wsmsm mmqu mm Mm mm 5 quotanyquot Wenti ca on Highest Fanout o Wrappers hide the hete e enhance scalability in info integration systems bination 0 Each heuristic fails on so 7 Cluttered sparse mixed 0 Heuristics fail on different web s1 s 0 Combine independent heuristics bas their probability for success Object ator Experiments ng RB 088 Si 073 SE 089 o 25 Web Sites 1P 0k R31 089 1076 pages SP 085 RgP 090 SIP 080 Test Data KSB 3991 R1 084 11ng 092 Algorithm Success Rate RSI 084 SIPB 092 SD 066 RIB 084 IPB 92 RP 075 SPB 087 s13 03 IPS 068 RP 088 RIPE 092 PP 079 Rs 088 RSPB 095 SE 088 R1P 088 RSIPB 095 1B 088 PB 088 RPB 088 68 0 Setup 7 25 Initial sites 1076 pages 7 50 Validation sites 2200 pages 0 Compute success rate for each algo thm test data 7 use results to calculate combination 0 Verify combination result on the validati 11 sites Object Initial Data Validation Data Algorithm Success Rate A ithm Success Rate SD 066 SD 069 RP 075 RP 086 IPS 068 IPS 0 PP 079 PP Q89 SE 088 SE 0 Combo 095 Combo 095 25 Web Sites 1076 pages 2200 pages 50 Web Sites Overview of IXP24OO Architecture Intel 2nd generation network processor designed for network processing goal to optimize for fast path and accommodate slow path programmable fast deployment of new services upgrades protocols needs external control processor WW OaEio w otn Bill hmmammmmm IXP24OO chip highlights 8 microEngines uE RISC 600MHz processors with instruction sets targeting networkingrelated operations for fast path XScale core standard embedded CPU ARM family running Linux slow path Media and Switch Fabric MSF Interface to RXTX packets from physical layer andor switch fabric ie main data path interface separate 32b RX and TX busses up to 125MHX W 8kB buffers various con gs up to 4prs ours 3XleEthernet connects to external devices and is configurable for different bus protocols DRAM controller for packet storage up to 2GB but access slow 300 cycles 64bit Wide SRAM controller store packet queues lookup tables 150 cycles 32bit Scratchpad memory 16k on chip for parameters communication bW uEs PCI bus 64bit66MHz connect to host external control processor or other PC1 compliant devices e g chain IXPs together Other units Hash control registers other peripherals timers interrupt controller perf monitors IXP2400 M Ev2 MEv2 John Morgan presentation at IXA University Summit04 Microengine MEVZ control store for instructions 4k 40bit instruction m uE 8 hardware supported threads hide memory access separate register sets PC fast ctX switch registers 256 GPR 32thd 512 Xfer 4X16thd DSXRW 128 NN 16thd addressing context relative or absolute GPRs local memory 640x32bits CAM 16 entry cache MicroEngine V2 lglbli El l5 Mom No l l Local 128 8 Memory Xfer In 7 640 words 128 D Xfer In 6 1 LM Addr LM Addr 0 r 0 PRandom F Glaxgl39rgmjrgl Eim lf CRC Unit Multiply I l F df Wt 32bit Execution 5 if 39quot quots 39 Data Path H Add shift logical w Other H Local CSRs ll Enm IPC messaging Via circular queues rings scratchSRAM operations for ring management 0 signaling 15 signals per context context can poll a signal explicitly use to guard critical sections Media Fabric Receive Logic Switch Fabric Unit Discarded idhpadm ILAB setup wwwcercsgatecheduproj ectsnpgilab hardware manual nethp31ixpdeVsdk3 0r 41DocsIXP2400 software tools Tools our IXP2400 Boards Radisys ENP2611 nethp3 1 iXpdeV CD Images IXAiEducati0n7Wkstn3 1 Docs Emma Sev a w 39 my gang 1 13 mama av WEE gtmmamw m i mw r m9mm amw mwagmgm WEBB IXP Software Development Two issues How to structuredesign your applications and How to actually conduct the development programming Programming model 11m 111 JJIHIIjJI 11m issues size of f1 cyclecount for f1 state across packets at f1 state for a packet across all s Microengine Programming Model Dispatch Loop Hardware abstraction OSlike functionality APIs for memory manipulations sram scratch dram bufferqueue management MSF access Protocol library header eld extraction validation update for popular protocols ipV4 Ethernet Utility library functions for hash table CAM accesses threads API Infrastructure library set up application speci c packet meta data pipeline parameters Software framework optlinksiXpIXA dataplanelibrary microblockslibrary EDUWkstn sampleapplication Radisys stuff SDKTools srclibrabry or metools Programming the IXPs Rely on tools Windows Software Development Kit Development Workbench Simulator Transactor cycle accurate simulation or so traf c generator build DLLs to implement interaction with special components eg custom protocols interactions with XScale or host Architecture Development Tool initial design and analysis nethp3 liXpdevexports sdk40SDKnassaupr8noncryptozip Program in microC familiar portable or microcode efficient best use of platform features or miX of both Build and run code with workbench tools hardware mode or microcode assembler and linker on ilab and XScale command line utilities on ilabiXp ssh ilabn gt telnet ilabnixpl gt load start stop sram CS48038803 ENC soon CS43656365 nterprise Computing unof cial TA Lenin Si garavelu 0 Write updates to database as the 0 Problems with recovery gtNeed to undo the aborted transaction op ations gt What if the system crashes between the decision to commit and the writing of last updates gtNeed to redo the committed but unfinished operations 0 Failure mode analysis gtMachine crash transaction abort 0 Machine crash gtDisk writes are NOT atomic gtThe Idea force write the information to di k before making any changes to support redo 0 Transaction abort gt Programmer may want to abort voluntarily gtUse the written information to undo Write ad Logging 0 Intentions List gtLampson and Sturgis 1976 unp gtPut all writes on stable storage intent39 ns 39 gtlf abort forget the list gtExecute the update operations at commit gtlf crash reexecute the whole list gt Write operations are idempotent gtReexecution repeats until successful completion gt Then erase the intentions list Number LSN conceptually a tim tam 0 Log contents gt Operations logging intention list gt Value logging old value amp new value 0 Operations Supported gtAppend during normal processing gtRead during recovery 0 Access Interface to Logs gt SQL Tables 0 Physical implementation gt Sequential duplexed files gtLog anchor the pointer to consistent data 0 Log Sequence Numbers gtEach log record has a unique id gtMonotonic increasing LSNs 0 Program Outline gtAcquire the log lock find file pa gtFill in log record update the anchor o LogMgr Daemon gtKeeps the log file open gtEncapsulates the actual lO operations 0 Log Flush Daemon gt Flush requests gtPeriodic timer interrupts 0 Serial write gtWrite one copy of the log gt When complete write other copy 0 PingPong gt Write page i and then page il gt When page full switch to new page gtParallel writing of both log files 0 Sometimes replicated logs Optimizations 0 Logging is a bottleneck gtDisk 10 is slow compared to CP gt Sequential logging is faster than ran in gt Steal and no force next slide 0 Group commit gtAmortize commit log ush gtDelay transaction commit until full use of log Page gtBatch only when there is high traffic 9 Pages 0 Stealing pages before com gtPush pages to disk optimistically gtlf commit you win gtlf abort need to undo the aborted value 0 Forcing pages at commit time gtPush pages to disk at commit time gtNo force leave pages in memory gtPush pages only when convenient gtlf system crashes must redo the committed value Optimizations 0 Rotational Delay gt Dedicated cylinders gt Write the first block under disk head 0 Multiple Logs gtlncreased lO bandwidth and complexity 0 Saving the Log Anchor gt Write the anchor from time to time gt Search for log end from the anchor gtPingPong write several anchors 0 DO program gtWrite a log record gtPerform the operation 0 UNDO program gtRollback the operation from appropriate 10 record gtEither by operation undo or old value 0 REDO program gtReperform the operation from log gtEither by operation redo or new value Res 0 Find the Anchor gtAccording to the writing method 0 Find Log End gtAccording to the writing method gt Sweep through good pages gt Stop when bad pages appear 0 Program Outline gt Reconstruct log anchor gtUndoRedo pending transactions gt Restart log daemons gt Start TP processing 0 It s possible to start TP processing early gtNeed to isolate the restart in a transaction gt Start normal transactions at restart lock point when the restart is guaranteed to succeed Trans 39on Manager 0 Basic Functions gtRecovery glue for centralized TP gt Coordinator for distributed transactio co gt Coordinates distributed recovery 0 Application Interface gtBegin T CommitT Abort T gt Savepoint Rollback gtPrepareT for 2PC gt ChainT for chained gtLeaveT and ResumeT for nested 0 TP monitor restarts gtLogMgr LockMgr TM and 0 TM Restart gtREDO committed transactions gt Find checkpoint record gt Start ResourceMgr restart gt Coordinate the end of all involved RMs 0 ResourceMgr Restart gtInitialization open files Identify gt Self recovery from private logs gt Coordinate with TM REDOUNDO 0 Independent RMs gtUNDO REDO scan local 0 System pairs for availability gtduplicate system somewhere else 0 Primary and a hotstandby gtBackup only partially useful 0 Assumptions gtNo network partitions gtReplicate programs network data 88031 F 02 Software Concepts and Techniques for Electronic Commerce mounency Control CS8803l Software Concept Techniques for ECommerc Calton Pu College of Computing Wren Transactions 0 Transaction Properties 7 Atomicity all or nothing 7 Consistency if updating DB preserve 7 Isolation isolate effects of concurrent tra 7 Durability committed updates survive failu 0 Sequence of data manipulation operations 7 Original ReadWrite 7 SQL Insert Delete Select Update Calton Pu Georgia Tech College of Computing Page 1 88031 F 02 Software Concepts and Techniques for Electronic Commerce we 0 Transaction A o Withdraw 50 7 Read account 7 Result 80 7 Result 80 7 Subtract 50 7 Subtract 20 7 Write 30 7 Write 60 o If A writes last then 0 If B writes last the result is 30 result is 60 0 Bank lost 20 0 Bank lost 50 Wrency Control 0 First Law of CC 7 Concurrent execution should be t 7 Sequential programs should work 0 Performance requirements 7 Low overhead 7 High concurrency OS solutions such as monitors won t do 0 Correctness criterion Calton Pu Georgia Tech College of Computing Page 2 88031 F 02 Software Concepts and Techniques for Electronic Commerce a izability 0 Sequential Execution 7 Satisfies ACID properties 7 OK for uniprocessor but limits scala 0 Concurrent Execution 7 lnterleaving actions overlap lost updates 7 Correctness serializability SR 7 SR execution that produces result equivalent a serial execution 7 Intuition of SR nondeterminism 7 CC algorithms that maintain SR TwoPhase Locking 0 Wellformed locks 7 Obtain an appropriate lock before 7 Read shared Write exclusive 0 Twophase locking 7 Acquire all locks before releasing any 7 Lock point in possession of all locks o Serializability 7 Timestamp of lock points induces a total ordering Calton Pu Georgia Tech College of Computing Page 3 88031 F 02 Software Concepts and Techniques for Electronic Commerce Mummy Control 0 First Law of CC 7 Concurrent execution should be t ie application programmers write program 0 Performance requirements 7 Low overhead per transaction 7 High concurrency beyond serial execution 0 Correctness criterion 7 Must make sense 7 Formal definition required today 0 Transaction Model 7 Execution history of RW actions 7 Serializable histories 0 Transaction Dependencies 7 Input action read 7 Output action write 7 Action dependencies Calton Pu Georgia Tech College of Computing Page 4 88031 F 02 Software Concepts and Techniques for Electronic Commerce WW Description 0 Dependency Graph 7 Data ow through the actions 7 Collapse the actions into a transaction 7 Transaction dependency graph R80 W wency Cycles 0 Cycle ltgt nonSR 7 Acyclic dependency ltgt partial or 7 partial order ltgt total order 7 total order ltgt SR 0 Kinds of Cycles 7 Lost updates see below 7 Dirty read 7 Unrepeatable read Overwrite Calton Pu Georgia Tech College of Computing Page 5 88031 F 02 Software Concepts and Techniques for Electronic Commerce o Locks 7 Read Shared WriteExclusive 7 Lock compatibility matrix cking 0 Well formed locks 7 Obtain appropriate locks before access 7 Unlock after use se Locking 0 No new locks after first un 7 Growing phase acquire locks 7 Shrinking phase release locks 0 Lock point 7 Before shrink holding all locks 7 Separates Gphase from S phase 7 Gives the serialization order Calton Pu Georgia Tech College of Computing Page 6 88031 F 02 Software Concepts and Techniques for Electronic Commerce Intuitio of Proof 0 2PL guarantees serializabi 1 7 Dependencies defined by the tim point 7 Sequence of lock point timestamps giV serial equivalent execution 0 Idea of the proof 7 Holding all locks force any dependent transactions to wait for this one to release lock they follow 7 2PL also forces previous lock holders to precede 0 Lock Manager interface 7 Lock names 7 Unlock names 0 Structure of Lock Manager 7 Multiple threads 7 Protected data structures 7 Encapsulated interfaces Calton Pu Georgia Tech College of Computing Page 7 88031 F 02 Software Concepts and Techniques for Electronic Commerce sm39 0 Acquire all locks and hol ct 2PL 7 Does not release locks incremen 7 Release all locks at the end of transa 0 Tradeoffs compared to generic 2PL 7 Less concurrency hold locks longer 7 Easier to recover avoids cascaded aborts 7 Easier to implement Locki 9 Logic 0 Lock is free 7 Insert into the lock hash table en 7 Link into the transaction table 0 Lock is not free 7 If compatible shared similar to above 7 If not compatible wait with care 0 Unlock 7 Delete this request from both tables 7 Check change grant locks or not 7 Wake up new requesters if granting Calton Pu Georgia Tech College of Computing Page 8 88031 F 02 Software Concepts and Techniques for Electronic Commerce 0 Many cases where locking delay others updates but to rea committed data cking 7 In the real world relaxed isolation is o adequate 7 Most of the time large part of DB always i committed state 7 Locking is useful would be useful to devise cheaper alternatives 0 Degree 0 atomic writes 7 Wellformed wrt writes 0 Degree 1 No lost updates 7 Wellformed wrt writes 7 Twophased wrt exclusive locks 0 Degree 2 no lost updates or dirty reads 7 Wellformed wrt writes and reads 7 Twophased wrt exclusive locks 0 Degree 3 ACID transactions Calton Pu Georgia Tech College of Computing Page 9 88031 F 02 Software Concepts and Techniques for Electronic Commerce mysis of 2PL 0 Most practical 7 Almost all commercial DBMS s 0 Performance advantages 4 Relatively low overhead in CPU 4 Good concurrency for low contention 4 Wellunderstood optimizations 0 Problems 6 Deadlock handling 6 Degradation at high data contention BaSIcltZimestamps 0 Static CC vs dynamic CC 7 Basic timestamps are assigned at of each transaction 7 That ordering is maintained as the seria 0 Each data access is checked 7 Past accesses are recorded reads are update 7 New access is allowed only if the timestamp ordering is legal Calton Pu Georgia Tech College of Computing Page 10 33 Crash Recovery CS8803l ware Concepts and 0 Write updates to database as the 0 Problems With recovery gtNeed to undo the aborted transaction op gtWhat if the system crashes between the decision to commit and the writing of last updates gtNeed to redo the committed but unfinished operations mole ance Basics 0 Failure mode analysis gtMachine crash transaction abort 0 Machine crash gtDisk writes are NOT atomic gt The Idea force write the information to di before making any changes to support redo 0 Transaction abort gtProgrammer may want to abort voluntarily gtUse the written information to undo Mb ad Logging 0 Intentions List gtLampson and Sturgis 1976 unp gtPut all writes on stable storage intent gtlf abort forget the list gtExecute the update operations at commit gtlf crash reexecute the whole list gtWrite operations are idempotent gtReexecution repeats until successful completion gt Then erase the intentions list Da base Log 0 Physical organization gtLog is a sequential linear file gtEach log record identified by a L0 Number LSN conceptually a tim 0 Log contents gt Operations logging intention list gt Value logging old value amp new value 0 Operations Supported gtAppend during normal processing gtRead during recovery 0 Access Interface to Logs gt SQL Tables 0 Physical implementation gt Sequential duplexed files gtLog anchor the pointer to consistent data 0 Log Sequence Numbers gtEach log record has a unique id gtMonotonic increasing LSNs tog Insert 0 Program Outline gtAcquire the log lock find file pa gtFill in log record update the anchor o LogMgr Daemon gtKeeps the log file open gtEncapsulates the actual lO operations 0 Log Flush Daemon gt Flush requests gtPeriodic timer interrupts We 0 Serial write ul Write gtWrite one copy of the log gtWhen complete write other copy 0 PingPong gtWrite page i and then page il gtWhen page full switch to new page gtParallel writing of both log files 0 Sometimes replicated logs moo Optimizations 0 Logging is a bottleneck gtDisk 10 is slow compared to CP gt Sequential logging is faster than ran gt Steal and no force next slide 0 Group commit gtAmortize commit log ush gtDelay transaction commit until full use of log Page gtBatch only when there is high traffic Bu eni 9 Pages 0 Stealing pages before com gtPush pages to disk optimistically gtlf commit you win gtlf abort need to undo the aborted value 0 Forcing pages at commit time gtPush pages to disk at commit time gtNo force leave pages in memory gtPush pages only when convenient gtlf system crashes must redo the committed value m 0 Rotational Delay Optimizations gt Dedicated cylinders gt Write the first block under disk head 0 Multiple Logs gtlncreased lO bandwidth and complexity 0 Saving the Log Anchor gt Write the anchor from time to time gt Search for log end from the anchor gtPingPong write several anchors Do 0 DO program gtWrite a log record gtPerform the operation 0 UNDO program gtRollback the operation from appropriate lo record gtEither by operation undo or old value 0 REDO program gtReperform the operation from log gtEither by operation redo or new value Resta Using Log 0 Find the Anchor gtAccording to the writing method 0 Find Log End gtAccording to the writing method gt Sweep through good pages gt Stop when bad pages appear 0 Program Outline gtReconst1uct log anchor gtUndoRedo pending transactions gtRestart log daemons gt Start TP processing 0 It s possible to start TP processing early gtNeed to isolate the restart in a transaction gt Start normal transactions at restart lock point when the restart is guaranteed to succeed Won Manager 0 Basic Functions gtRecovery glue for centralized TP gt Coordinator for distributed transactio gt Coordinates distributed recovery 0 Application Interface gtBegin T CommitT Abort T gt Savepoint Rollback gtPrepareT for 2PC gt ChainT for chained gtLeaveT and ResumeT for nested 0 TP monitor restarts gtLogMgr LockMgr TM and 0 TM Restart gtREDO committed transactions gtFind checkpoint record gt Start ResourceMgr restart gtCoordinate the end of all involved RMs Databa e Restart 0 ResourceMgr Restart gtInitialization open files Identify gt Self recovery from private logs gt Coordinate with TM REDOUNDO 0 Independent RMs gtUNDO REDO scan local Disaster Recovery 0 System pairs for availability gtduplicate system somewhere else 0 Primary and a hotstandby gtBackup only partially useful 0 Assumptions gtNo network partitions gtReplicate programs network data 88031 F 02 Software Concepts and Techniques for Electronic Commerce 1B Basic E Architecture CS8803l Software Concept Techniques for ECommerc Calton Pu College of Computing Commu ications 0 Message passing circa 1978 7 Formatting messages decoding mes 7 Waiting for messages resend if lost e 0 Remote Procedure Calls 1983 SOSP 7 Remote services just like local services 7 Automated packingunpacking of parameters 0 HTTP circa 1990 7 Back to the future Calton Pu Georgia Tech College of Computing Page 1 88031 F 02 Software Concepts and Techniques for Electronic Commerce Theory and ctice of TP 0 Theoretical background 7 Serializability 7 Concurrency control 7 Recovery algorithms 0 Practical algorithms 7 Twophase locking 7 Writeahead logging 0 TP Monitor architecture 0 How do TP Monitors address clien 0 TP Monitors vs DB Servers 0 Application Servers Calton Pu Georgia Tech College of Computing Page 2 88031 F 02 Software Concepts and Techniques for Electronic Commerce Work w 0 Extended transaction models 0 Work ow managers 0 Advanced transactional applications 0 Continual Queries Standing q 0 XWrap Access to heterogeneous o Infosphere Information ow 0 Ubiquitous Computing Aware Home 0 PeertoPeer PeerCQ PeerTrust Calton Pu Georgia Tech College of Computing Page 3 88031 F 02 Software Concepts and Techniques for Electronic Commerce m Topics 0 Security privacy trust 0 Digital cash micropayments 0 Standard interfaces from EDI 0 Types of ecommerce B2B B2C C2C 3Tier Clien Server Today Web Server Calton Pu Georgia Tech College of Computing Page 4 DllUlL LUulbU llallbdbllUll flUbUbblllg 14A CS8803I Software Concep Techniques for ECommerc Calton Pu College of Computing Napste 0 Shawn Fanning Northwestern 7 Estimated 60M users by 2000 7 Shut down in 2001 0 Client server implementation 7 Index is centralized 7 Immutable files decentralized replicated in unreliable nodes P2P claim The DataIntensive Systems Center DllUlL LUulbU llallbdbllUll flUbUbblllg Gnutell 0 AOL aquires Nullsoft 7 Justin Frankel and Tom Pepper writ original software called Gnutella 7 Gnutella is posted for one day 200003 0 Gnutella reverseengineered and refined 7 Pure PeertoPeer no servers Gnutella He Sharing 0 Nodes are called servems 7 Both client and servers 0 Each Gnutella instance will 7 Store selected files 7 Route queries file searches from and to its neighboring peers 7 Serve file if file stored locally The DataIntensive Systems Center DllUlL LUulbU llallbdbllUll flUbUbblllg Search 0 Gnutella request by A creates 7 Search String S 7 Unique Request ID N 7 TimetoLive T and hops passed 0 Check local system if not found 7 Sends A S N T to all Gnutella neighbors 7 Flooding is not scalable 0 If you don t have the le you 7 Query 7 of your partners 7 If they don t have it they contact 7 of t partners for a maximum hop count of 10 0 Stop ooding by checking 7 Unique ID stop when seen twice 7 Hop count exceeded The DataIntensive Systems Center DllUlL LUulbU llallbdbllUll flUbUbblllg o B Receives Gnutella request A S 7 If B has already received request N or T request and does nothing a B looks up S locally and sends N Result 0 If not found locally 7 B sends B S N T 1 to all of its Gnutella neighb and it records the fact that A has made the request N a When B receives a response of the form N Result from one of its neighbors it forwards this response to A Group 0 Join with a PING to announce 7 Receivers forward the PING to neig 7 Receivers backpropagate a PONG to a self IP address numbersize of shared fil 0 Periodic refresher of network state 7 PING again 7 Wellknown root nodes if starting from scratc The DataIntensive Systems Center OllUlL qulDC 1 lallbabLlUll I lUbCDDlllg ma a Network Size Nodes m he largest network component 200 l Figure I Network growthThe number ofnodes in the largest con nected component in the network grew by about Z5 times over the seven months ofour study N n a 8 Messages per second 0 74 oitu a Do suia vas h Nut m 22 NuNg m Trme Immutes Figure 2 Generated tmmc In November 2000 overhead traf c on a randomly chosen link accounted for more than 50 percent ofthe total whereas user query messages were only 36 percent The DataIntensive Systems Center ouUl I JUul ab 1 lallbabLlUll L lUbU331115 Some tatistics 0 Network size growth in 7 mon 7 From 2K to 48K nodes 0 Network instability 7 About 40 of nodes live less then 4 hours 7 About 25 of nodes live more than 24 hour 0 Message types in June 2001 7 About 91 queries 7 About 8 PING messages New rk Diameter Reachable node pairs percent l0 ll ll 1 2 3 4 5 s 7 s 9 Node7to7node shonest path hops Figure 3 Node tonode shortest paths More than 95 percent of node pairs could be reached within 7 hops The DataIntensive Systems Center QilUl L JULll DC 1 idllbaleUll l iuqumuE PowerLaw o Selforganizing networks follo werlaw 7 Nodes with L links is proportional 7 Many nodes with a few links 7 A few nodes with many links 0 Gnutella network seems more resilient 7 Nodes have more than a few links Monnectivity Nodes log scale iOU Links log smie Figure 639 Connectivity distribution March May 200 IThere are too few nodes with low connectivity to form a pure powerlaw nelwor M The DataIntensive Systems Center OllUl L bU Lll DC 1 lanoap LlUll I lULCbblllg Overlay Net rk Mapping o Figure 7 Mapping the overlay network topology to the network infrastructure a Mth perfect mapping a message inserted into the network by nodeA travels physical link DE only once to reach all other nodes b With inef cient mapping the same message traverses the link six times 15 Random Clus Er Hypothesis 0 Gnutella is insensitive to physi 7 Gnutella clusters appear randomly c 7 De ne entropy function to match hypot random choice different domain names 7 Same domain names indicate low entropy 0 Good clustering reduces entropy 7 Gnutella clustering does not reduce entropy The DataIntensive Systems Center Introduction to Enterprise Computing 9A Dell Case Stud and ECommerce Instructor Calton Pu unofficial TA Lenin Singaravelu Slides Credit Deli Case Competing in the Network Era for IT Value in the Netcentric Organization Kenneth L Kraemer University of California Irvine Calton Pu Georgia Tech College of Computing Page 1 Introduction to Enterprise Computing Order model to undercut competitor prices 0 Dell starts at the bottom of each market moves up eg low end servers to higher end servers 0 lntemet and IT a key component of Dell s strategy Dell and what problems they are having IMarket segmentation Tailor products and ervice 70 Corporate 20 SMB 10 government an education Value to corporate IS departments Complete PC outsourcing from sales through disposit on amp replacement Reduce PC ownership costs help them support customers Calton Pu Georgia Tech College of Computing Page 2 Introduction to Enterpnse Computing Direct relationship with customer is strat 39c rich information flows Outsource nonstrategic functions Information flows substitute for physical flows stomer relations Third party HW and sw suppliers Internet based customer services Online ordering asset tracking product road ma highly tailored information all provided online Changed name from Dell Computerto Dell Inc De sells PCs servers printers storage networking so network management services Sels addon software peripherals PDAs cases cameras TV Promote Dell as company that knows how E worksquot Run s Dell on Dell corporate customers come to Dell for advice Dell has eservices business in partnership with consulting firms g Accenture to capitalize on Dell s reputation as ecommerce leader Results Dell now 1 PC vendor with 31 in US and 18 globally Revenue of 35 billion growth over 30 annually Calton Pu Georgia Tech College of Computing Page 3 Introduction to Enterprise Computing Co 39ng the virtual company with IT a enetworks Speed Orderdriven processes linked by internal only 7 hours of inventory in factory and orders to Entire value network linked by EDI Internet extran Quality Bar coding allows components to be tracked to suppliers occur stop production and notify suppliers Cell assembly allows problems to be fixed on the spot without down production Cost Online sales and support saves on call center costs Supply chain coordination substitutes information for inventory Results SGampA overhead 8 compared to 15 for others 1 09 inventory turns annually minimizes depreciation New technologies can be introduced immediately Dell IT as for coordination I Customer I d external networks allow 39lled in 5 days or less l Dell Online Web browser interface Message broker software I Other applications I I Dell Order Management System I J Finance Production Software Shipping Service installation Support 8 l Telesales Calton Pu Georgia Tech College of Computing Page 4 Introduction to Enterprise Computing 0 Have to discount old technology 0 Retailer sends unsold units back to manufac rer 0 Strengths of the direct model 0 Most ef cient method of distribution 0 Extremely low inventories 0 Rapid response to customer changes 0 Strong relationship with customers and suppliers Results 0 Far 0 Direct vendor Dell has best pe 0 Dell has forced other vendors to g di 0 Industry as a Whole has improved performance Calton Pu Georgia Tech College of Computing Page 5 Introduction to Enterpnse Computing US PC et Share 1990 200 O Appe Compaq I DeH Gateway HP iBM US PC market share Operational performance Dell amp industry Calton Pu Georgia Tech College of Computing Page 6 Introduction to Enterprise Computing Industry improvement inventory turns Inventory Turnover for PC rms 2003 versus 1999 number of turns 1999 2003 Dell 60 109 Gateway 35 36 Apple 12 116 IBM PC Division 14 43 Compaq PC divisionsa 15 HP Personal Systems Groupb 18 Industry 2 92 0 Direct model amp network org made for the Internet 0 Works well with individual amp corporate 0 Works well with suppliers and business pa 0 Works well on a global scale 0 Network model of organization exhibits superior performance 0 Dell has outperformed the industry on all measures 0 Other rms unable to catch up 0 Dell continues to gain market share at their expense14 Calton Pu Georgia Tech College of Computing Page 7 Introduction to Enterprise Computing 0 CRM Customer Relationship 7 Frontend B2C 0 SCM Supply Chain Management 7 Backend B2B 0 ERP Enterprise Resource Planning 0 EAI Enterprise Applications Integration 0 AppServers Web Servers DBservers etc 0 Start from a precise model of b activities human and machine 7 Explicit and verifiable properties dependencies and constraints 0 Gradually decompose into lower levels 0 abstraction 7 Generate executable code easy part that observes the dependencies and constraints Calton Pu Georgia Tech College of Computing Page 8 Introduction to Enteipiise Computing Media Sho xample Increas Maikel share Consult cataiague Buy Menu Items Quallly Packages Mean Suppllsr Fig 1 i Model for a Media Shop commumg Euslness Tro Specification Dependency Continuous Supply ype goa lVIode 111ai11tain Deponder l quot Depondee Media Supplier Attribute constant item Mctliz39iltum Ful llment condition for dependcr lmy BuyItcmlJust C39reatedbuy buyil 77LillSlOClCl the media retailer 0X1qu to gut items in StuCk as mun as someone is interested in buying them c r C 3 Fig 2 librsz Tmpm Speciiicutions Calton Pu Georgia Tech College of Computing Page 9 Introduction to Enterprise Computing Means Ends Analisys Comm nlcauun Servlces Increase Market Shar Buy Medl Items Media Shop l Canginuing ausmess 4 Handle aquot a S e s r ers x a r 1 I S Satisfy Cuslomer Dsires nepenuerixeeuepemee Actor Dependency Goa Task r W mm D Cg r39 19 Legend Ressuwte 50mm Adar mmdaly Availability lnternet Telecom Services pr B rowse c atalogue A Process onnne Money ransacuons Keyword Search F d User New Needs Place Order a 05919J18 Security Process Intern at O rders lap V Aouepue n Adaptability KJ Buy Med a Items Increase Market Share Con nuing Busmess Media Shop M d e la Supplier Media Items Ha oustgr rs Calton Pu Georgia Tech College of Computing Page 10 Introduction to Enterprise Computing uo a Ol eleJls Lepo Calton Pu Georgia Tech College of Computing Page 11 Introduction to Enterprise Computing More Spe 39 ications TELL CLASS StructureInSMetaClass IN Class WITH C1ass is here used as a MetaMetaClass attribute name String part exclusivePart dependentPart ApexMetaClass Class CoordinationNetaClass Class MiddleAgencyMetaClass Class SupportMetaClass Class OperationalCoreMetaClass Class END StructurelnSMetaClass Fig 8 Sti ucture in S in Telos ination Actor TELL CLASS CoordinationMetaclass IN Class WITH Class is here used as a MetaMetaClassV attribute name String taskDepended s StandardizeTask WITH depender OperationalCoreMetaClass Class ND goalDepended c ControlGoal WITH depender MiddleAgencyMetaClass Clas s END softgoalDepender s StrategicManagementSoftGoal WITH dependee ApexMetaClas s 1135 5 END END Coordinat ionMetaclass Calton Pu Georgia Tech College of Computing Page 12 Introduction to Enterprise Computing Principal FannaLn Aulnariiy Delagatlon Jmnl Management Joint Venture iza 39on Comractual Agreemeni Principal Fannerj su l in Second nillees 69 Resource Exchange anclpal Panneriz Knowledge Sharing 1 2 4 4 1 Predictability Flat Structure Women Sti llctllrc illS 3 A I I 39I39 Pvmmhl 4 i lelnlVI HTHTC l Blillllllg 39Ihlwvcr Al39lil s Lougth Hicrcllicul Contracting 11391 lL39Ill llllL39gl39HllUu 7 m iptiil ion 7 7 2 Calton Pu Georgia Tech College of Computing Page 13 Introduction to Enterprise Computing Corre 39nTabIeCont d 7 5 39 5 CooperatiVity J Flat Structure Structure in 5 Pyminiil 7 ilointJ39Qnth o Bidding T ulteovor Arm s Length Aggr ability Hiei cllicnl Contracting Vertical Integration Coop tution Adaptabilin KWHHNVWVW Secur y raf V e 7 I Claim r Claim WWW Con m l V Pussibie Cuui39hcts Dynamic y x run 53 m Eon 1 x t y We CW1 V r em sen Amheiuicnticn Con uentimiuy K ly Idemy man i Van micquot K A rm 39 r in E ensihility M figgiiity Upd ll J lity m e r J lt1 Claim External Agents can a uire trusts information mm Wis pymmid J uvemm Completion Fig 10 Selecting the Architecture Calton Pu Georgia Tech College of Computing Page 14 Introduction to Enterprise Computing Custqmer Profller n den Q 39 iality 39 Order Processor Accounnm mixing Processor I 29 Example 0 ents 1 F r Locate Provtder Requested Service Provider AdveFt39se Serwce F1 12 Matdnnacker Calton Pu Georgia Tech College of Computing Page 15 Introduction to Enterprise Computing Item ser Interactions Statistics Processor U epresentation r actor ltlt39 actor goal dependency Fwd Source Change task dependency No Change My goal dependency care r39 actor Source Match Monrtor ltltI39 actor goal epenoency 7 Route Into Request ask depeHCency gt guaIdependency Prewoe lnrormatron Prome Customer V r actor 39 3m OnLine Inr o Searcher task dependency HIIs Information Catalogue ask dependency 4quot I task dependency TransIate Response A Query Infonnatron Source Mediator Wrapper I actor task de endenw Ask Ior Info Advenlsrng i A Calton Pu Georgia Tech College of Computing Page 16 w I gtgt ltlti actorgtgt I an or CartForm ShoppingCan CustomerProfIler ltlt39 ado ltltTextgtgt ttemCounl Integer echum tmeger ltltTexlgtgt qty0 integer ax currency on39lme ltltTextgtgt currentTolal currency g fmg y Catalogue ltltCheckboxgtgt selectttemo 39 Vu ltlt mm C 9 km l l gorillagllllggl oglngcl rrency ltltSubmltgtgtA dlt n 1 nte39er 0quot CustomerPIo leCard Subm Con rm gaDlmlms39m Earcwemy r a r ltltButtongtgt ancel Medlaltem CUSOme rung 8 gtgtR H ItemCounN nolmcattom cuslomerN me smug U 0 90am cal cullattegitalso quot5 0quot9 lvrstName 5t ge CanU ca cua eWyOl1 t n br Ing nuddleName string bulldnemmme wimLDU 99 mmrm ItemTllle51 ng cd m 99 me em I r1 5 meTame 0 lmuahzaRepmo ItemBarCode OLE e sung pdmenemso Plans 7 r ItemPIcture O edgllalld aslgmg loadCartForm Inmallze rem allegory 31mg rcfessnmstnn updatetartFormU Selecmem Propose genre 5mquot p 39 g ktllcartForr addltem SUCCSded descrlplton slnng I A l sa ary in ger M 0 checkout remDVEl em edttorstrtn manlalStaIus stung cance con rm D pu Sher string famrlyCompw I integer Sl mem mr IOQOUI lame te date rnternetPreqo 10oolean Immune U veiliycg I d unrtPrrce currency no un ers oo entheSPrfgogll lislrtng ge aemDe a s weight single 0 Ies u sun 1 comments 5 n 1ng 0 ltemDetatl credilcardn Inle er ally m egberb I prequrchase G D39 a 0W5 u 5 ocean 39slnn e lit prevPLuchPrIce039D 391 30 0 Integer Introduction to Enterprise Computing Fig 3916 Partial Cltho Diagram for Stow me Focusing on Shopping Cut2 t h k t W processOrder D D r 3 Pl r 39 mung mlormauon 1 v actorgtgt ltltr actorgtgt Detwe stetrstrcs Processor Pro cessor r payment request j r r idehver rteta process rnvorce checkoutrrequest for proposal Tlmeout tZISD 9 31 53 FIPA contract Net Protocot nolrunderstood Noll callon customer Shopping can 1 a at 53 checkuutrrfp r 2 emse nolrundersmud prupose cancerproposat c r cancelrproposal U a eptrpraposal succeeded failure r r Decision acceptrproposal D Succeeded t inform failure Plan Dragram cf next gure Calton Pu Georgia Tech College of Computing Page 17 Introduction to Enterprise Computing Checkout 4 Mandaldry elds lled V r lverlfyCC ccrx vall Press orl unon elds Cred Cam lconnrrngl ScollHrnm Hem Checklng Checkrng Regrslerlng Not con rmed F5 cancelo cc ndl valr Ndl all rnandam A K31 elds lled A K5 l 1dreacn ems selected llem Reg s em Flelds Updated Updated Flnal Amounts Calaulatlng l Records updated Stock Records Updatlng Amounts Calculated Tl w cuslonEr Dlsplaylng succeeded R990 Q Cueldmer Prorrle Updatlllg shopplngCan logou repdrl asked rd ll l lmllallzeReporlO Bpgaled Q Already regrslered g g unleaqu 595 Shopplngcan lnmallzeo From Specifi on to Code Proc checkO uShnppillng shapCart lt slmpC art failalsmpquotal39f Irngn SloijrngCnl sllnpquotar1 gt lt predeedC l39lllccBurton a reiniiulnaS39IluppillgC39ul slopC39lu39t gt H lt Timwmn gt Elli 7 6inifialiurSlnppingC m tshan39nl l gt gt shopC al39f Actil39atEdC39lzeckoutButton F I39essellquotl k0lltButtoll sllzl39lC39llcCl39OuMslLIpCuI39l gt EndProc F 19 ConGolog like Speci cation for the Checkout plan Calton Pu Georgia Tech College of Computing Page 18 Introduction to Enterprise Computing intentjrsrqe neg 7 ii39 39 39 ntengj essde ned quotquot A V AA g x WV A Dependency A eeded ngoax satls es Actor Euigu r ng quot L M 39 5 quot 7v V r depender H epe39n39de x39quot I r x m V BD Agequot perceives gt Belief BIOLEggiquot Desire V achleves j t quotrlh ai erlvVquot reelizeda 5057 handled as mapped m re fmzdea posted a planned as 7 quot 39 mmms hd 39u JACK 39 b dsi mw 4 quotA r Jack Agent 95 P J399I gt DE relation thug i BDGoaEvem 4 i reads Tm0 1 e aggregated mto quot 39 r 97593 aggregatedlnto apablg of aggregated Into quotrib capablllty Fig 20 EDIJACK mapping overview 4 JACK Release 20 I J Fils Edit Windows Help Project Media 31 PrupErti651Madie gurus 363 Agents extends Plan Showinncm 39 mamas nntifitatinn 4 extends Aganl nnlund2rsnud I I I E Databases 4 extends Plan 35 brwata Tlmeout 5 handia nntifitatinn capabilities Epmpnsa Events extends Plan 59 Pom quot ti m quot 39 handias nomimtmn ans 3 Event 585 fuse I mum44 uses Immm menus MussagEEvEnl e 1 uses nuLundzrsluud a pmmg Memo uses mum L5 vuid clued qule i l Glimmer a mni minn EWEMS 9M1 I Extends MnssagEEvnnl E ems a Pasting Methods 59 quotm5 WWW rquot Ly vnid nnti mtinno j canabnines y g Hans 35 Datghases hg chntknul Ehu mm Extends CIDSEdWIIrId A extends Plan g handles checkoutJfP 5 EMS key lung ShuppingCanjd E Immune 38 Calton Pu Georgia Tech College of Computing Page 19
Are you sure you want to buy this material for
You're already Subscribed!
Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'