960 likes | 1.25k Views
IEEE ICDE ‘98 Tutorial: Mobile Computing and Databases. Margaret H. Dunham Southern Methodist University Dept of Computer Science and Engineering Dallas, Texas 75275 mhd@seas.smu.edu http://www.seas.smu.edu/~mhd. Outline. Introduction & Data Management Issues Query Processing Caching
E N D
IEEE ICDE ‘98 Tutorial:Mobile Computing and Databases Margaret H. Dunham Southern Methodist University Dept of Computer Science and Engineering Dallas, Texas 75275 mhd@seas.smu.edu http://www.seas.smu.edu/~mhd
Outline • Introduction & Data Management Issues • Query Processing • Caching • Data Broadcasting • Transaction Processing • Agents • Projects & Products • Conclusion ICDE/SMU - Dunham
Mobile Computing Architecture ICDE/SMU - Dunham
Terminology • Fixed Network (FN) • Base Station (BS) (Mobile Support Station - (MSS)) • Fixed Hosts (FH) • Cell - Area covered by BS (1-2 miles) • Handoff - Changing BS by intercell move • Mobile Host (MH) (Mobile Unit (MU)) ICDE/SMU - Dunham
Wireless Networks • Cellular • High Cost • Scalability Issue • Limited Bandwidth: 10 Kbps • Wireless LAN • Traditional LANs with wireless interface • Low Cost • Limited range: 10-100 meters • Bandwidth: 10Mbps • NCR Wavelan, Motorola ALTAIR ICDE/SMU - Dunham
Wireless Networks (cont’d) • Satellite Services • Wide Coverage • Very Expensive • Low Bandwidth: 1-2Mbps • Paging Networks • Wide Coverage • Sky Tel, Motorola • Slow: (Ethernet: 10Mbps; FDDI or switched Ethernet: 100Mbps; ATM: 155Mbps) ICDE/SMU - Dunham
Handoff • Changing BS due to movement between cells • State information transferred • Current handoffs in cellular phones may take up to a few seconds with breaks in conversation of 100-300 ms. • Soft - Temporarily connected to two BSs • Hard - Only connected to one BS ICDE/SMU - Dunham
Location Management • Tracking mobile user • User associated with home location server (Home Agent) • May augment by searching in local area first • May augment with user profiles • Mobile IP [11,14] • Triangle Routing • Route Optimization • Location Control (Routing Agent) S Ah Af M ICDE/SMU - Dunham
Location Management (cont’d) • Active Badge (Cambridge,[2]) • Track employees and route telephone calls • Unique code emitted every 15 seconds • Sensors placed in offices and corridors • Location Information Replications • No HLR • Hierarchy of Location Servers • Each server maintains information about its subtree ICDE/SMU - Dunham
Mobile Applications • Information Services (Yellow Pages) • Law Enforcement and Medical Emergencies • Sales and Mobile Offices • Weather, Traffic, Sports, Entertainment • Trucking • Cellular Subscribers in the United States: • 90,000 in 1984;4.4 million in 1990;13 million in 1994 • Handheld computer market will grow to $1.77 billion by 2002 ICDE/SMU - Dunham
Technology Push • Internet: ftp, telnet, email, http,html • Advancing Wireless Communication Technologies • Laptop, Notebook, and Palmtop Computers ICDE/SMU - Dunham
Classification of Mobile Database Systems ICDE/SMU - Dunham
Data Management Issues • Speed of wireless link • Scalability • Mobility • Location dependent data; Location specific queries • Limited by battery power • Disconnection (Voluntary, Involuntary) • Replication/Caching • Handoff ICDE/SMU - Dunham
Insurance Example ICDE/SMU - Dunham
Medical Example • 911 Call • Ambulance arrives/departs • Closest hospital • Access patient records • Send vital signs • Update patient records • Page hospital personnel • Order medical supplies ICDE/SMU - Dunham
MC/DB Research • Transaction Processing • Caching - Replication • Broadcast Disks • Agents • Mobility • Location Dependent Data • Recovery • ACID (?) ICDE/SMU - Dunham
Outline • Introduction & Data Management Issues • Query ProcessingLocation Dependent Queries and DataNew Query TypesQuery Optimization • Caching • Data Broadcasting • Transaction Processing • Agents • Projects & Products • Conclusion ICDE/SMU - Dunham
Location Dependent Data • Value of data depends on location • Temporal Replication - One consistent value at one time • Spatial Replication - Multiple different correct data values at one time • Temporal Consistency - All data objects satisfy a given set of integrity constraints. • Spatial Consistency - Consistency constraints satisfied within Data Region. • SMU/University of Missouri at Kansas City, [17] ICDE/SMU - Dunham
Location Dependent Queries • Result depends on location • Different from traditional distributed goal of location independence • Ex: Yellow Pages, Directions, Map • Predicates based on location: “Find the cheapest hotel in Dallas.” • Location constraints: “Find the nearest hotel (to me).” ICDE/SMU - Dunham
Similarity to Spatial Queries • Spatial Data: Data associated with space occupied by object. • Types of spatial queries: contains, contained in, intersects, neighboring, east of, etc. • Spatial data structures • Spatial operators • Spatial selects and joins • PSQL - extend SQL, [18,20] ICDE/SMU - Dunham
Differences from Spatial Queries • Client is actually moving • Location of client may be part of the query itself • May depend on direction of movement • Data may not directly contain location information • Includes temporal features as well Spatial data is dynamic ICDE/SMU - Dunham
Querying Moving Objects • Moving Objects Spatio-Temporal (MOST) data model • Dynamic Attributes - Change over time • Queries over temporal history: • Instantaneous - Ex: “Find all restaurants I’ll reach in the next half hour. ” • Continuous - Ex: “Find all restaurants within 5 miles.” The answer continuously changes as the MU moves. • Persistent - Ex: “Find the cars that travel greater than 10 miles in the next half hour.” • Future Temporal Logic (FTL) language • University of Illinois, [20] ICDE/SMU - Dunham
Query Optimization • How best to satisfy the information request made by the client? • Different Cost Factors: I/O, network • Different Access Options: cache, FN, broadcast • Dynamic and Adaptable - environment changes • Alternative plans include deciding (based on state of MH and environment) whether to access in the cache at the MH, to request a mobile transaction, or to obtain from a broadcast disk. ICDE/SMU - Dunham
Outline • Introduction & Data Management Issues • Query Processing • CachingOverviewTypesResearch • Data Broadcasting • Transaction Processing • Agents • Projects & Products • Conclusion ICDE/SMU - Dunham
Caching • Placing data at MU. Usually on disk. • Faster to access from MU than from DBMS in fixed network. • Facilitates disconnected operation. • Adaptive to connection mode. • Not just another replica • Pull based • Most work on files not databases ICDE/SMU - Dunham
Caching Functions • Data fetching • Granularity (Page, file, table, semantic) • Replacement • Coherency • Callback - Servers send invalidation messages to clients. • Detection - Clients send queries to servers to. • Updating during disconnect • Data integration when reconnected ICDE/SMU - Dunham
Connectivity and File Systems Table 3.2 from [15] ICDE/SMU - Dunham
MU Replica Control Protocols • Traditional Replication Protocol problems: • May hinder mobility • Quorum Consensus: Can’t get quorum if disconnected; Avoid using MU replicas to make up quorum • Location information not always readily available • Primary Copy: Should not be stored at MU • First class/Second class replicas ICDE/SMU - Dunham
Checkpointing Table 3.4 from [15] ICDE/SMU - Dunham
Prefetching vs.. Hoarding • Both prefetch data in anticipation of future use. • Prefetching • Objective is to improve performance (throughput or response time). • Cache miss not catastrophic. • Hoarding • Objective is to fetch all needed data into MU cache prior to disconnect. Thus the goal is to facilitate disconnected operation. • Cache miss is catastrophic. • OK to overfetch ICDE/SMU - Dunham
Hoarding/Spying • Listening to and recording file accesses • Performed during a snapshot interval • May be combined with user profiles. • Results limited to the snapshot. ICDE/SMU - Dunham
Disconnected Issues Table 3.1 from [15] ICDE/SMU - Dunham
Coda • First project to demonstrate disconnected operation. • Optimistic Locking • Granularity - sets of files. • Coherency - callbacks • Hoard Walking: Periodically (every 10 min) evaluate contents of cache. Recalculate priorities. • On a callback break, object is purged, refetching on demand or during next hoard walk. • Venus - cache manager at MU ICDE/SMU - Dunham
Coda (cont’d) • Venus states: hoarding,emulating,write disconnected (earlier reintegrating). • Cache misses during disconnection are treated as failures. • During disconnection, a log (Change Modify Log) of operations is created. Hoarding Strong Connection Weak Connection Disconnection Write Disconnected Connection Disconnection Emulating ICDE/SMU - Dunham Adapted from Fig 2 in [34]
Coda (cont’d) • During integration, log applied. Conflicting updates are determined and user assists in resolution • Timestamps at volume and object level used to determine conflicts. • Trickle Reintegration used to asynchronously propagate updates. • Hoard Profile - list of files and priorities. • Lowest priority objects chosen for replacement. • Weak Connectivity - low bandwidth, high latency, high cost, or intermittent • CMU, [29,34] ICDE/SMU - Dunham
Little Work • Disconnected AFS • Cache operations depends on type of connection • Connected - Continuous; High bandwidth; Normal operation • Partially Connected - Continuous; Dialup; Delayed writes • Fetch Only - On demand; Cellular; Optimistic replication • Disconnected - Fail if cache miss ICDE/SMU - Dunham
Little Work (cont’d) • Caches 64KB chunks of files • Fetch only mode • Modifications sent back to primary file server • Conflicts stored separately and user notified • Michigan, [25,26] ICDE/SMU - Dunham
Seer • Ficus • Uses semantic information to determine contents of cache. • Semantic distance between files measured in number of file accesses on average between two files. • Access is defined as open-close. • Distance measure used to cluster files. Fetching of a cluster based on user hints and LRU information. • UCLA, [24,30] ICDE/SMU - Dunham
Summary Table 3.1 from [15] ICDE/SMU - Dunham
Sleepers and Workaholics • Cache invalidation report • Periodically (synchronous) the server broadcasts report of changed data. • MU waits for next report prior to answering query. • Sleepers - frequently disconnected; cache invalidation based on signatures. • Workaholics - rarely disconnected; periodic broadcast of changes. • False Invalidation • MITL and Rutgers, [21] ICDE/SMU - Dunham
Transparent Analytic Spying A B C D E F Access Tree • File Working Sets • Continually observe and record (in a log) file access. At hoard time, reference the log to determine hoard. • Trees for a process are created reflecting file access pattern. One tree per program execution is generated. ICDE/SMU - Dunham
Transparent Analytic Spying (cont’d) • Hoard all files or only those in in the most recent execution. • Tracing adds about 2% CPU overhead. Average space for file log record is 100 bytes. • Implemented on Unix, NFS, Mach • Cache miss rate over wireless slightly higher than on wired. • Prefetching overall reduced cache misses and elapsed time • Columbia, [36] ICDE/SMU - Dunham
Predictive File Caching • Analyze file access patterns in different environments: Personal productivity, Programming, Commercial • Working Set Statistics: Mean working-set sizes small (18MB per day) • Attention Shift Statistics: 0.6 per user per week • Conflict Statistics: Depends on environment • Conclusion: • Hoarding is possible due to small working set size • LRU caching insufficient • UCLA, [31] ICDE/SMU - Dunham
Virtual Primary Copy • Mobile Primary Copy (MPC) at MU • Virtual Primary Copy (VPC) at BS • Global transactions access VPC • Consistency of VPC maintained by BS • BS monitors MU disconnect • Multilayered approach is transparent to other sites • Monash, [23] ICDE/SMU - Dunham
Roam • MC Replication System • MU Peer to Peer communication allowed • Ward Model: • Ward: Grouping of replicas for locations that frequently communicate • Ward Set: Set of replicated data stored in a ward. • Ward Master: Doorway into ward. Maintains consistency between wards. ICDE/SMU - Dunham
Roam (cont’d) • Ward members are “close” • No “pre-motion” operations • Intra-ward synchronization easier than inter-ward • Reconciliation- Synchronization process • Selective replication at file level • Scales well • UCLA, [33] ICDE/SMU - Dunham
Semantic Cache • Caching granularity at a predicate level • SPJ query - Materialized view • Advantages: reduces network overhead, reduces cache space • Disadvantages: Indexing, query trimming • Semantic Cache - C = {Si} • Semantic Segment - Si=<Sr,Sa,Sp,Sc> • SMU ICDE/SMU - Dunham
Outline • Introduction & Data Management Issues • Query Processing • Caching • Data BroadcastingOverviewIndexingResearch • Transaction Processing • Agents • Projects & Products • Conclusion ICDE/SMU - Dunham
Data Broadcasting • Server continually broadcasts data to MUs. • Scalability: Cost does not depend on number of users listening. • Mobile Unit may/may not have cache. • Facilitates data access during disconnected periods. • Allows location dependent data access. • No need to predict with 100% accuracy the future data needs. • Broadcast based on probability of access. • Periodic broadcasting of all data. ICDE/SMU - Dunham
Data Broadcasting (cont’d) • Classification: • Coverage - Everything, Subset • Content - Static, Dynamic • Indices - Index, Self Descriptive • Data Stream - Flat, Skewed, Multiple Disks • Client - Passive, Active • For uniform page access, flat disk has best expected performance. • With skewed page access, nonflat disks are better. • Push based. ICDE/SMU - Dunham