Replication, Caching, Prefetching and Hoarding for Mobile Computing

1
Replication, Caching, Prefetching and Hoarding
for Mobile Computing
2
Definitions

• Replication To maintain multiple (consistent)
  copies of a data item
• Static replication the number and location of
  copies are statically determined (at compile
  time, design time).
• Dynamic replication the number and location of
  copies is determined dynamically (at run-time)
• Caching To maintain a temporary copy of the data
  in fast (local) memory. The copy is fetched when
  it is first accessed.
• Pre-fetching To obtain a temporary before it is
  accessed (to hide access latency).
• Hoarding To preload a copy of a data object so
  that the mobile client can work while it is
  disconnected from the network (I.e. prefetching
  to tolerate disconnections).

3
Data Access Model

• On-Demand
• Broadcast Channel

4
Motivation

• Caching (Prefetching/Hoarding) at mobile clients
  is crucial to improve performance of info access
  and database querying.
• Issues
• Read only data currency guarantees
• Read/Write data consistency in presence of
  disconnected operations
• Server Load/Scalability in presence of numerous
  clients

5
Mobile Database Querying Requirements

• Minimize query delay
• Maximize number of queries answered per unit time
  (system throughput)
• Handle client disconnection
• Conserve wireless bandwidth and battery power
• Minimize server load
• Handle mobility

6
Advantages of Caching in Mobile Environment

• Helps reduce latency caused by narrow bandwidth
  wireless links
• Enable limited functionality in mobile hosts even
  in disconnected mode
• Helps conserve battery power by reducing the
  number of uplink queries
• Conserves bandwidth

7
Problems in Maintaining Consistent Cache

• Classic solutions do not work
• Mobile Clients may be disconnected for long
  duration gt invalidations may be lost
• Upon reconnection mobile clients will have to
  revalidate their cache (wastes energy and
  bandwidth).
• Need new solutions

8
Challenges to Efficient Caching Scheme

• Efficient caching scheme should take into
  account
• Data access pattern
• Data update rate
• Communication/access cost
• Mobility pattern of the clients
• Connectivity characteristics
• Disconnection frequency
• Available bandwidth
• Data currency requirements
• Location-dependence of information

9
General Issues in Designing Caching Schemes

• Where to cache? How many levels of caching to
  use?
• What to cache (when to cache a data item and how
  long) ?
• How to invalidate cached items? Who is
  responsible for invalidations? What is the
  granularity at which the invalidations are done?
• What data currency guarantees the system can
  provide to the user? What are the costs involved?
  How to charge the user?
• What is the effect of the caching scheme on the
  query delay (response time) and the system
  throughput (query completion rate)?

10
Classification of Cache Invalidation Schemes

• Who is in charge of invalidations?
• Server or Client (Push or Pull) Callbacks or
  Validation Checks
• Whether or not server maintains per client state
  information?
• Stateless or Stateful Server
• How server sends invalidation reports?
• Synchronously or Asynchronously
• What kind of information is sent in the
  invalidation report?
• State or History based
• How information is organized in invalidation
  reports?
• Uncompressed or Compressed

11
Cache Maintenance Schemes

• Broadcasting Invalidation Reports Barbara Sigmod
  94.
• Disconnected Operation in CODA (Satyanarayanan
  et. al. )
• Hoarding (Prefetching)
• AS (Asynchronous Stateful) Caching Scheme (Kahol
  et. al. ICDCS 00)

12
Broadcasting Invalidation Reports

• Uses stateless servers and synchronous broadcasts
  Barbara Sigmod 94
• Clients maintain local caches and use the
  information in invalidation reports to update
  their cache.
• A server broadcasts invalidation reports every L
  time units which contains ids of all the data
  items which changed during the past w kL time
  units.
• A query is satisfies after receiving the next
  invalidation report.

13
Broadcasting IR Variations

• If a client is disconnected from the network and
  misses k consecutive invalidation reports then it
  has to discard its cache.
• Two variations
• Timestamp Strategy (TS) invalidation reports
  contain ids of modified data items over a large
  window (k gt 1).
• Amnesic Terminal (AT) invalidation reports
  contain ids of only those data items which
  changed since the last broadcast (k1).
• TS is better when clients are sleepers and AT
  is better when clients are workaholics.

14
Disconnected Operation in CODA

• Goal COnstant Data Availability
• Mechanisms server replication and disconnected
  operations.
• Caching scheme (asynchronous, stateful)
• Uses callbacks while a client is reachable from a
  server.
• During disconnections permits access to possibly
  stale data.
• Upon reconnection, the client does validity
  checks on each volume cached.
• Uses hoarding to improve data availability

15
Drawbacks

• Drawbacks of Barbaras scheme
• Poor delay characteristics due to waiting
  involved before answering a query.
• Poor network utilization characteristics due to
  answering of queries in bursts.
• Does not support arbitrary disconnection pattern.
• Drawbacks of CODA caching scheme
• Server has to keep cache state of each client
  (affects scalability).
• A client has to perform volume-by-volume
  validation check after each reconnection.

16
AS Caching Scheme (Kahol et al)

• Maintains a Home Location Cache (HLC) at home MSS
  of a mobile client.
• A HLC contains the state of the cache at a MH.
• Uses Asynchronous transfer of invalidation
  reports.
• Supports arbitrary disconnection durations by
  maintaining the timestamp of the last
  invalidation report destined for an MH at its
  HLC.

17
An Example for AS Scheme

• Each cache is associated with a cache timestamp
  which is the timestamp of the last invalidation
  report received.
• A mobile client sends a probe message to its home
  MSS when it gets connected to determine whether
  it missed any invalidation reports while it was
  disconnected.

18
Hoarding

• Planned and accidental disconnections are not
  considered failures.
• A technique to reduce the cost of cache misses
  during disconnection
• load necessary data before disconnect and be
  ready.
• Hoarding techniques
• user-provided information (client-initiated
  disconnection)
• explicitly specify which data (files, tables) to
  hoard
• Implicitly based on the specified application
• access structured-based (use past history)
• E.g., tree-based in file systems, access paths
  (joins) in databases

19
Hoarding versus Prefetching

• Both pre-fetch data in anticipation of future
  use.
• Prefetching
• Objective is to improve performance (throughput
  or response time).
• Cache miss is 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

20
Hoarding in Database Systems

• Granularity of Hoarding
• RDBMS ranges from tables, set of tables, whole
  relations
• OO DBMS objects, set of objects or class
• Hoard by issuing queries or materialized views
• User may explicit issue hoarding queries
• E.g., Create View with Update-On clause Lauzac
  98
• OO query to describe hoarding profiles
  Gruber 94
• History of past references both queries and data
  objects
• Hoard Keys - an extended database organization
  Badrinath 98
• hoard keys are used to partition a relation in
  disjoint logical horizontal fragments

21
References

• D. Barbara and T. Imielinski, Sleepers and
  Workaholics Caching Strategies in Mobile
  Environments, VLDB Journal, 4, 567-602, 1995.
• A. Kahol, S. Khurana, S.K. S. Gupta, and P. K.
  Srimani, A Strategy to Manage Cache Consistency
  in a Disconnected Distributed Environment, IEEE
  Transactions on Parallel and Distributed Systems,
  12(7), 686-700, July 2001