Design and Performance Studies of an Adaptive Cache Retrieval Scheme in a Mobile Computing Environment

1
Design and Performance Studies of an Adaptive
Cache Retrieval Scheme in a Mobile Computing
Environment

• Paper by Wen-Chih Peng and Ming-Syan Chen
• Presented by Arseny Bogomolov
• October 25, 2005

2
Presentation Outline

• Setting the Context
• Problem Statement
• Related Work
• Preliminaries
• Cache Retrieval Models
• Cost Analysis for an Adaptive Scheme
• Performance Study
• Conclusions

3
The Typical Mobile Environment

• A traveling mobile user (MU)
• A running application
• Communication with application server
• MU moves to a new service area in a distributed
  server architecture
• Service Handoff
• Minimizes communication overhead
• Minimizes application delays
• Balances workload
• Fault tolerance

lt- Must be seamless!
4
An Example Application

• Traffic Reports
• Shortest distance routes
• Up-to-date traffic status
• Location-dependent information
• Live traffic video streaming wirelessly
• Temporal Locality?
• What and where can be cached?

5
System Architecture

• SA Application Server
• LBn Local Buffer
• One per MU
• Caches data
• For one Ui
• Coordinator
• Concurrency control
• Transaction monitoring
• Can also cache data
• For Ui..n

6
The Problem A Mobile User

• The user changes service areas
• Service Handoff must occur

• LBA has cache for Ui, LBB is empty

Cache
7
The Proposed Solution

• Three caching schemes
• From Local Buffer (FLB)
• From Coordination Buffer (FCB)
• From Previous Server (FPS)
• Which one to use?
• Depends on transaction properties
• Temporal Locality
• Cost of cache miss
• Authors propose DAR

8
Dynamic Adaptive Cache Retrieval Scheme (DAR)

• Authors main objective contribution
• A set of decision rules for service handoff
• Selects most appropriate of the 3 schemes
• For each phase of transaction
• Based on transaction properties

9
Paper Contributions Summary

• Evaluates properties of FLB, FCB, and FPS
• Proposes DAR scheme for service handoffs
• Comparative analysis of the 4 schemes
• Simulations to validate the results

10
Related Work

• Caching at the proxy servers
• Service handoff not considered
• Cache invalidation schemes
• Impact of client disconnects on performance
• Energy-efficient schemes
• Adaptive schemes adjust size of report
• For the most part all schemes are static
• Previous use of coordinator
• Concurrency control
• Transaction execution monitoring
• Not used for caching

11
Preliminaries

• Temporal Locality tendency of data pages to be
  referenced again soon
• Intra-transaction Locality same data pages
  referenced in same transaction
• Inter-transaction Locality same data pages
  referenced by consecutive transactions
• Type of temporal locality influences type of
  caching scheme

12
From Local Buffer

• One per each MU
• Created when MU
• enters service area
• First transaction triggers warm-up
• No pre-fetching is done
• Best when no temporal locality
• Find shortest distance to X from where I am
• Saves on pre-fetching costs

13
From Previous Server

• Data is fetched from
• previous server
• Best for intra-transaction locality

• User is in service area of SA
• Find all routes to home (pages 23, 26)

• User moves to service area of SB

• Pages 23, 26 referenced twice

14
From Coordinator Buffer

• Data fetched from
• central location
• Best for inter-
• transaction locality

• MU Uj is in Sc
• Frequent updates of parking info (44, 39)

• MU Ui moves to service area of SB
• Still wants up-to-date parking info
• Pre-fetched data results in a cache hit

15
Dynamic Adaptive Retrieval

• Need to pick best scheme for each stage based on
  transaction properties
• Initial Stage
• Consider FLB and FCB
• Execution Stage
• FLB, FCB, or FPS?
• Termination
• Coordinator buffer write

Service Handoff
16
Cost Analysis and DAR

• Want to derive rules for picking best scheme
• Whats considered?
• Number of cache entries per user - N
• Cost of cache miss - Cm
• Cost of cache replacement
• Transaction properties (temporal locality)
• Pintra probability page has intra-transaction
  temporal locality
• Pinter probability page has inter-transaction
  temporal locality
• PCB probability page w/temporal locality is in
  CB

17
DAR Rules for Initial Phase

• FCB or FLB?
• Cache miss costs
• FLB N Cm
• N (Pintra Pinter)
• N (Pintra Pinter) (1 PCB) -gt Ntl
• N (1 (Pintra Pinter)) -gt N-tl
• FCB (Ntl N-tl) 2Cm

Data w/temporal locality
CB misses for temporal
CB misses for non-temporal
18
Rules for Initial Phase (contd)

• FCB profitable if
• Cost of FCB lt Cost of FLB
• (Ntl N-tl) 2Cm lt N Cm
• (Pintra Pinter) gt 0.5 / PCB
• If temporal locality is prominent, FCB is used
  (CB has more frequently used pages)
• If it is not, use FLB to avoid unnecessary
  pre-fetching costs

19
DAR Rules for Execution Phase

• FCB, FLB, or FPS?
• If no temporal locality, go with FLB (same as for
  Initial Phase)
• Otherwise, decide based on ratio of intra- to
  inter-
• Use threshold f to decide
• Pintra/Pinter f ? FPS
• Pintra/Pinter lt f ? FCB

20
DAR Rules Summary

• Initial Phase
• Execution Phase

21
Performance Study

• 64 servers - 8x8 mesh topology
• MU moves randomly between servers
• TXNSIZE of data objects (4K page) in
  transaction (20-30)
• Cache sizes
• Local Buffer 1 of DBSIZE
• Coordinator Buffer variant
• DB has K data objects
• Pintra K (intra-transaction objects)
• Pinter K (inter-transaction objects)
• Ppseudo K (objects w/no temporal locality)

22
Impact of Temporal Locality

• Pinter fixed at 10, Pintra varied
• Pintra fixed at 10, Pinter varied

(Pintra Pinter) gt 0.5 / PCB 0.5 / PCB 0.1
23
Performance of DAR

• Ppseudo set at 20
• Vary Pintra/Pinter ratio

24
Performance of DAR (contd)

• Can adapt to transaction properties
• Higher cache hit ratio regardless of whether MUs
  transactions have more inter- or intra- pages
• Employs the appropriate cache retrieval method
  dynamically
• Pintra/Pinter f ? FPS
• Pintra/Pinter lt f ? FCB
• What should the value of f be?

25
Performance of DAR (contd)
1.60
1.75
2.25
CB 20 DBSIZE
CB 50 DBSIZE
CB 80 DBSIZE
26
Performance of DAR (contd)

• f is at the point of intersection of FCB and FPS
  cache hit ratio curves
• With size of coordinator buffer increasing, it is
  expected that hit ratio of FCB will increase
• Therefore, f increases
• But f decreases if we increase Ppseudo
• In the end, value depends on estimation of Pintra
  and Pinter

27
Paper Summary

• Examines several cache retrieval schemes to
  improve performance during service handoff
• Presents a cost model for evaluating the schemes
  presented
• Analyzes impact of temporal locality
• Analyzes impact of coordinator buffer size
• Performance analysis of all schemes presented,
  including validation on simulator

28
Conclusion

• FPS best for intra-transaction locality
• FCB best for inter-transaction locality
• Adaptive algorithm (DAR) presented

• Higher cache hit ratio of inter-transaction
  pages in transactions with prevalent
  inter-transaction pages
• Higher cache hit ratio of intra-transaction
  pages in transactions with prevalent
  intra-transaction pages
• Makes DAR advantageous over other schemes