GeoGrid: A scalable Location Service Network

Authors: J.Zhang, G.Zhang, L.Liu

Georgia Institute of Technology

presented by

Olga Weiss

Com S 587x, Fall 2007

OverviewIntroduction

Problem addressedWhat is GeoGrid?Existing solutions review

Design of GeoGridBasic GeoGrid systemConstructionRoutingDual peer and Dynamic work load adaptation techniques

Experimental results and conclusion

IntroductionProblem addressed

serving a large and growing number of mobile userscontinious delivery and dissemination of location-based information in real time

GeoGrid – a geographical location service overlay network system

DecentralizedGeographical location awareTechniques to improve fault-tolerance and workload balance

Existing solutions reviewCurrent location-based services are

constrained to fixed set of moving objectsexpensive to maintain and expand

One approachCreate and maintain a centralized graphical location service. Drawbacks:

Response timeExpensive access to infrastructure communication serviceNot robustNo model for such large scale location-based service

ChallengesCan we organize nodes into efficient service network that is geographical proximity aware?End-to-end communication b/w any two nodes is boundedHow to handle workload imbalance (hot spots)? How to minimize the possible service interruption?

Ex. Highway system can be heavily loaded during the rush hours

GeoGrid DesignBasic GeoGrid systemGeoGrid constructionRoutingDual peer and Dynamic work load adaptation techniques

Basic GeoGridA network of N nodes interconnected using GeoGrid topology and routing protocolA node is a point in 2dim geographical coordinate spaceSpace is dynamically partitioned into N disjoint rectangleEach node “owns” a rectangular region

Basic GeoGridNodes self-organize into an overlay networkConnectivity is established through immediate neighbors of a nodeA mobile user connects his mobile device to one of the nodesEach node runs GeoGrid middleware and serves as a proxy for the mobile user

Basic GeoGridAssumptions

Information services existence (provide geographical info)User can be either from outside of the network or from inside itNetwork nodes are not mobile

GeoGrid ConstructionPlane is divided among N nodes into a set of rectangular regions r = <x,y,width,height>A node p is identified by a tuple <x,y,IP, port,properties>Each node maintains a list of its neighbors

GeoGrid ConstructionGeoGrid is constructed incrementallyStart from one node owning the entire GeoGrid spaceSplit the space upon new nodes joining

decide which region the new node q belongs (say, of node p)split the region in halfhand one half to node qnotify neighbors (they must add q into their lists)create a list for q from the list of p

GeoGrid ConstructionBasic bootstrapping process for a new node

Obtain geographical coordinate from a serviceObtain a list of existing nodes from a bootstrapping serverRandomly chose an entry node from the listInitiate a joining request contacting to the entry node

Routing in GeoGridLocation queryRequest is tagged with (x,y)

p issues a query (x,y); q = pif q does not own (x, y) then forward the request to a neighbor closest to (x,y)

Routing in GeoGridTwo critical issues:

Load balancingRouting efficiency

Solution:heuristic load balance scheme - workload dynamic adjusting

Dual peer techniqueLoad adaptation technique

Dual peer technique

Improves the overall system reliabilityMaps region sizes to the capacities of region owner nodes

2 nodes share a region ownershipPrimary owner node is a node with the larger capacitySecondary owner node is a backup node

Dual peer techniqueNode join:

First 3 steps are the same as in basic GeoGridChose a neighbor region that has one owner and the least capacityIf no such region exist, chose the region with the least primary node capacitySplit the region and becomes a primary owner node of the region

Node departureSecondary owner departure causes no changePrimary owner must inform neighbors

Dual peer techniqueFailure recover

Status information is periodically synchronized b/w primary and secondary nodes

Failure of a node:If the primary node leaves, the secondary node becomes the primaryIf the secondary node leaves, the region becomes half-fullIf both nodes die, what happens?

Dual peer techniqueAvantages

Improves the fault resilienceReduces the number of region split operationsImproves the system load balance

Dynamic workload adaptation

Main idea: balance workload distribution by selectively assigning new nodes to the most heavily loaded regions in its neighborhoodThree basic rules:

Use local adaptation instead remote one (less operation overhead)Use secondary peer switching/moving (for primary peer the ops costs more)Region split/merge are expensive – should be used with less priority

ConclusionExperimental results show that GeoGrid can reduce the workload load imbalance by an order of magnitudeUnique design:

Use of geographical mapping of nodes to regions; improved routingDual peer and Dynamic workload adaptation techniques reduce load imbalance and improve fault-tolerance