E conomics 101 Demand and Supply Equilibrium (think load balancing) - PDF document

O utline � Introduction Food for thought Mariposa: a wide-area distributed � Economics 101 database system � Paper - background Previous assumptions � Key issues with assumptions � New requirements � M. Stonebraker et al. � Paper - approach Mariposa – model and architecture UC Berkeley � Business process � Presentation: Nikhil All other stuff (storage, naming…) Discussion: Albert � Results � � Epilogue I ntroduction E conomics 101 � Demand and Supply � Equilibrium (think load balancing) � What is D in D DBMS? � Requirements � Why do we care? � Some examples (90’s and beyond) http://www.sci.wsu.edu/math/Lessons/SupplyAndDemand/theory.html O utline P revious (wrong!) A ssumptions � Introduction Food for thought � Static data allocation � Economics 101 � Single administrative structure � Paper - background Previous assumptions � � Uniformity Key issues with assumptions � New requirements � � Paper - approach Mariposa – model and architecture � Business process � All other stuff (storage, naming…) � Results � � Epilogue 1

K ey I ssues (with assumptions) N ew R equirements � Static data allocation For non-uniform, multi-admin WAN � Is it distributed? � Scalability � Single administrative structure � Data mobility � No free will � No global synchronization � Uniformity � Total local autonomy � Surreal � Easily configurable policies F ood f or T hought O utline Food for thought � Introduction � Economics 101 Starter � Paper - background Previous assumptions � What applications can you imagine using � Key issues with assumptions � Mariposa for? New requirements � � Paper - approach � What about systems using the previous Mariposa – model and architecture � assumptions for distributed system? Business process � All other stuff (storage, naming…) � Results � � Epilogue M ariposa – T erminology M ariposa – E conomic M odel � Account � ABC of market economics � Budget � Budget allocation � Broker � Advertising service � Bid � Bidder � Join/leave system (fragment trading) � Rush � Buy/sell => bidding An Introduction to the Rush Language (A. Sah, J. Blow, and B. � Dennis) Unlike ‘if’ -> tests a predicate at one point � Like an ‘if’ which is always watching � on { $x < 10 } do { puts stdout $x } � � Stride (handicapped ||) 2

M ariposa – A rchitecture M ariposa – E conomic M odel � Client Retail Market Insti. Market Client 1 � Middleware Site 1 Exit Client 2 Site 2 � Local EE Broker Client 3 Site 3 Enter . . . . Site 4 Client n . . . . Site n M ariposa – A rchitecture M ariposa – A rchitecture � Client � Client � Middleware � Middleware � Local EE � Local EE M ariposa – A rchitecture M ariposa – B usiness P roc. Free market at its best � Client � Players (clients) willing to buy data at a � Middleware certain price � Players (bidders) willing to sell data at a � Local EE certain price � Neutral mediator (broker) to enforce fair-play � Broker � Expensive bid protocol � Purchase order protocol 3

M ariposa – B usiness P roc. F ood f or T hought Bidding � Bid acceptance Main Course � Search problem � Does this bidding process seem like a � Coalesce (in strides) and then evaluate � Heuristics (at last!) good model to you? Can you think of � Finding bidders other applications for which it might be � Good advertisement leads to higher sales ☺ relevant? � Setting bid price (auto load balancing) � Network bidder M ariposa – N aming M ariposa – S torage M gmt � Names � Idea => Maximize revenue � Internal � Fragment trading � Full � Splitting and coalescing � Common � Name contexts � Name resolution � Name discovery T ests and I ssues O utline � Introduction Food for thought � Economics 101 � Environment � Paper - background � 3 relations, 3 sites, 11 MB (!) of data Previous assumptions � � Purchase Order Vs Expensive bid Key issues with assumptions � � Bidding New requirements � � Issues � Paper - approach Mariposa – model and architecture � Market eco ^ AI => only simplistic � Business process � possibilities All other stuff (storage, naming…) � Results � � Epilogue 4

E pilogue F ood f or T hought � Mariposa – where is it now? Dessert Mariposa > Cohera > PeopleSoft > Oracle (But!) � � DDBMS – commercial and deployed � How does Mariposa compare to None that I know of � today's P2P's system? How is it the � DDBMS – current research same? How is it different? P2P systems � Semi-structured and Un-structured data � � What Can Databases Do for Peer-to-Peer? The state of the art in distributed query processing � Thank You 5