OnDemand Transient Data Backup in Mobile Systems

1
On-Demand Transient Data Backup in Mobile Systems

• Jeffrey Hemmes,
• Christian Poellabauer, and
• Douglas Thain
• University of Notre Dame

2
Introduction

• Problem managing transient data in
  human-portable mobile devices
• Examples Localized sensor data, configuration
  files, log files, etc.
• Many situations require only short-term backups
  not long-term persistent storage
• Swap batteries, replace device, etc.

3
Introduction

• Challenges in mobile networks
• Short battery life ? low availability
• Limited capacity (storage, CPU, network)
• Simple solution replication
• Where to replicate data?
• How much capacity is required?
• Traditional approaches can be expensive
• A lightweight approach is required

4
Assumptions

• Cooperating, trusted users with portable
  computing devices
• Mobile nodes capable of self-localization
• Approximate wireless radio range is known
• Relatively isotropic RF propagation model
• For simplicity, not necessarily required!

5
Resource Monitoring

• Determine storage requirement and current system
  state
• Periodically stat specified files and maintain
  running total of storage space needed
• Poll available free disk space and current
  battery state of local device
• Determine current physical location
• Advertise capabilities and location
• Use routing protocol or broadcast messaging

6

• Short-term off-device storage of transient
• data is needed now what?

7
Peer Selection

• Phase I eliminate unsuitable peers
• Routing protocol augmented with position and
  system state information
• When data backup is required
• Evaluate directly connected nodes
• Eliminate those peers with insufficient capacity
  from further consideration

8
Peer Selection

• Phase II estimate availability window
• Use location history of directly connected nodes
  that may potentially receive data
• Project availability based on speed, trajectory,
  and wireless radio range
• Use greedy selection
• Transfer data as needed

9
Node A
Node C
Batt 5 Disk 22M
Batt 66 Disk 20M
Node D
Node B
Node 0
Batt 87 Disk 20M
Batt 99 Disk 19M
Required 6 MB
10
15 s
Node A
Node C
Batt 5 Disk 22M
Batt 66 Disk 20M
118 s
36 s
Node D
Node B
Node 0
Batt 87 Disk 20M
Batt 99 Disk 19M
Required Space 6 MB
11
Node D
Node 0
Batt 87 Disk 20M
Required Space 6 MB
12
Simulation Setup

• 50 nodes in 300 meter wireless range
• Initial speeds and directions of peer nodes
  random with uniform distribution
• Motion patterns either linear or random
• Velocities for moving peers random values between
  1 and 4 meters per second

13
Evaluation
Linear Motion
Random
Nearest
Location Aware
Success Rate
Elapsed Time (s)
14
Evaluation
Random
Availability Time (s)
Nearest
Location Aware
Wireless Range (m)
15
Open Problems

• Performance
• Policy can be used to prevent pouncing
  bottlenecks, but at what cost to overall
  performance?
• Should lower priority data be purged or further
  offloaded to other devices?
• Recovery Semantics
• How to deal with migrating data?
• Security
• How to deal with authentication and access
  control?
• Garbage Collection
• What to do with unrecovered data?

16
Conclusions

• Mobility prediction has been used to improve link
  longevity we combine with system state to find
  remote storage space
• Greedy selection based on availability can
  increase recovery rates for short-term remote
  storage of transient data
• Approach particularly beneficial in larger-scale
  mobile networks

17
For More Information

• Cooperative Computing Lab
• http//www.nd.edu/ccl
• TeamTrak Mobile Computing Testbed
• http//www.nd.edu/teamtrak
• Jeffrey Hemmes
• jhemmes_at_nd.edu
• http//www.nd.edu/jhemmes

18
Acknowledgment

• This research is partially supported by DoD
• Defense University Research Instrumentation
• (DURIP) Grant Number W911NF-06-1-0120

19
Questions
20
Evaluation
Random Motion
Random
Nearest
Location Aware
Success Rate
Elapsed Time (s)
21
Evaluation
Random
Nearest
Location Aware
Success Rate
Storage Requirement (MB)