Download Slide 1

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Sensor Network Research Group
University of Massachusetts, Amherst
Ultra-Low Power Data Storage for Sensors
Gaurav Mathur, Peter Desnoyers, Deepak Ganesan and Prashant Shenoy
Motivation:
 Number of flash memory storage options exist for sensors
 NOR flash (Mica, Mica2, Telos)
 Multimedia cards – MMCs (RISE)
 NAND flash (UMass Mica2 adapter board)
 No comprehensive empirical comparison of the energy-efficiency of
storage alternatives for sensor devices
 Many applications require energy-efficient, high-capacity storage
 Camera sensing
 Acoustic sensing
What is the most energy-efficient storage platform for sensor devices ?
How does the energy cost of storage compare to that of computation and communication ?
What are the implications of an ultra-low power storage subsystem on sensor net design ?
Comparison Study of Storage Technologies
Toshiba Parallel NAND
flash adapter
Mica2 Atmel serial NOR
flash
TelosB STM
serial NOR flash
MMC adapter
 Compare MMC, NOR & NAND flash technology
 Measure active & sleep mode power consumption
Implications of Low Power Storage
 Challenges conventional wisdom of trade-offs
Computation < storage << communication
 Energy-efficiency of NAND flash enables ultralow power, almost infinite storage (~1 GB)
High-capacity, Low-power storage impacts the following
sensor application domains by supporting:
 Higher degree of local data archival and indexing, which
supports in-network query processing
 Use of history for efficient network-level compression
Comparison of the per byte cost of operations (in μJ)
Device
Mica2 NOR
TelosB NOR
Hitachi MMC
Parallel NAND
Read
0.26*
0.056*
0.144
0.004*
Write
4.3*
0.127*
0.659
0.009*
Erase
2.36
0.185
0.0033
0.004
Total
Impact on Radio Transmission Costs
6.92
0.368
0.806
0.017
* Without ECC. Cost of performing ECC in software is approx 0.026μJ/byte
 Parallel NAND flash is 21 times more efficient than
Telos STM flash and 407 times Mica2 Atmel flash.
Comparison of computation, communication and storage costs
TI MSP430
instruction
Energy
Ratio
0.0008
1
Toshiba
NAND read
0.004
5
Toshiba
CC2420
CC2420
NAND write Radio Tx Radio Rx
0.009
11
2.6
3250
2.2
2750
Storing data on NAND flash is 369 times
more energy efficient than communication
 BMAC uses a per-packet preamble that leads to a high
per-packet transmission cost
 Instead, store packets on NAND flash & batch transmit
 Can lead to a 58x reduction in communication costs !
http://sensors.cs.umass.edu/projects/essense