Incremental Network Programming for Wireless Sensors. IEEE SECON 2004 Jaein Jeong and David Culler UC Berkeley, EECS

Transcription

1 Incremental Network ming for Wireless Sensors IEEE SECON 2004 Jaein Jeong and David Culler UC Berkeley, EECS

2 Introduction Loading to Wireless Sensors In System ming Most Common. ming time is in proportion to # nodes. Network ming Sending whole code over radio still takes time. Code Host Machine Serial or parallel port In system programming Sensor nodes Code Host Machine Radio Channel Network programming Sensor node

3 Introduction Incremental Network ming source code is changed in small amounts. Sensor nodes Reduce programming time by sending the difference. Host Machine Radio Channel Code Version 1 + Difference Code Version 2 Incremental Network programming

4 Previous Work Single-hop Network ming: XNP Multi-hop network programming: MOAP, Deluge Extends the Range Incremental network programming: Reijers / Kapur Reduces the ming Time Virtual machine programming: Maté / Trickle Small Application Level Code Our incremental network programming approach Difference Generation using Rsync. Platform independent solution.

5 Network ming Steps Incremental Network ming (1) Encoding: Generates the difference. (2) Dissemination: Transmits the difference. (3) Decoding: Rebuilds the new code. Host Machine Sensor Node (1) Encode (2) Dissemination (3) Decode User app binary Dissemination Data Radio Packets External Flash User Application Section Boot loader Section Host Machine Sensor node

6 Design Considerations Reduce the amount of data transmission. Minimize the access to the external flash memory. Avoid expensive operations for sensor nodes.

7 Step 1: Difference Generation (Encoding) Storage Organization Memory Running External Flash Memory images for previous / current version Sensor Node Memory On-chip Memory User (L bytes) Boot loader SRAM Internal Flash External Flash L bytes Memory Previous L bytes New For others

8 Step 1: Difference Generation (Encoding) First Approach: Fixed Block Comparison Comparing at fixed sized blocks: Doesn t work when code is shifted. Previous New =? insert nbytes shift Remaining

9 Step 1: Difference Generation (Encoding) First Approach: Fixed Block Comparison Comparing at every byte: Finds shared blocks with high cost. Need an efficient way of finding shared blocks. Previous New =? B byte window For each byte Remaining

10 Step 1: Difference Generation How to Find Shared Blocks Efficiently? Two level checksums (Idea of Rsync algorithm) Finds a matching code block quickly with high accuracy. Checksum (1 st level): Fast but not accurate (32-bit). Hash (2 nd level): Not fast but very accurate (128-bit). Previous Hash Table (Checksum, Hash) New =? Checksum Hash B byte window For each byte Remaining

11 Step 1: Difference Generation Using Rsync Algorithm (1) Build hash table for previous image. (2) For the window of new image, calculate checksum. (3) Lookup hash table. For a matching checksum, calculate hash. Otherwise, move to the next byte and repeat (2). Previous Hash Table (Checksum, Hash) New (3) (2) (1) =? Checksum Hash B byte window For each byte Remaining

12 Step 2: Dissemination Modified Rsync Protocol for Wireless Sensors Modified Rsync protocol for resource constrained sensor nodes. Sender (3) Compares checksums & generate difference (1) Asks version of blocks (2) Sends checksum list (3a) Sends difference Receiver Original Rsync algorithm (4) Rebuilds the latest version Sender (1) Calculates the checksums. (2) Generates the difference (3) Sends the difference. Receiver Modified Rsync algorithm (4) Rebuilds the latest version

13 Step 3: Decoding Rebuild during Code Delivery Host sends difference as a sequence of messages. Sensor rebuilds new image using the diff. Optimizing Flash Memory Access Copy blocks are aligned to flash memory record boundary. Host Machine Sensor Node Difference Message Previous New Matching block Copy Non-maching block Download Download Copy

14 Experiment Setup Test Platforms: MICA2 / MICA2DOT Test Applications Simple network programmable app: XnpBlink and XnpCount. Test Steps Test app code and modified code are given to host program. Compare the xmit time with that of XNP (non-incremental). Test Application Modified Test Application + Dissemination Data Sensor nodes Host

15 Experiment Setup Test Cases Case 1: Changing a constant (XnpBlink) command result_t StdControl.start() { return call Timer.start(TIMER_REPEAT, 1000); } Case 2: Modifying implementation file (XnpCount) event result_t Xnp.NPX_DOWNLOAD_DONE (uint16_t wid, uint8_t bret, uint16_t weenofp){ if (bret == TRUE) call CntControl.start(); else call CntControl.stop(); return SUCCESS; } Case 3: Major change (XnpBlink XnpCount)

16 Experiment Setup Test Cases Case 4: Modifying configuration file (XnpCount) Commented out IntToLeds component. Case 5: Modifying configuration file (XnpCount) Commented out IntToRfm component. configuration XnpCount { } implementation { components Main, Counter, /*IntToLeds,*/ /*IntToRfm,*/ TimerC, XnpCountM, XnpC; } // Main.StdControl -> IntToLeds.StdControl; // IntToLeds <- Counter.IntOutput; // Main.StdControl -> IntToRfm.StdControl; // Counter.IntOuput -> IntToRfm;

17 Results Fixed Block Comparison: Almost no speedup except for case 1. Using Rsync algorithm: Speed up of 2 to 2.5 for a small change (case 2 and 4). Still limited speed up for big changes (case 3 and 5). Speed up Estimated Speed Up (Fixed) Measured Speed Up (Fixed) Estimated Speed Up (Rsync) Measured Speed Up (Rsync) Case 1 Case 2 Case 3 Case 4 Case5 Modifying Comment XnpBlink to Comment Comment Constant Lines XnpCount IntToLeds IntToRfm

18 Problems of Rebuild during Delivery Difference Delivery Still Not Optimal Difference is decoded without being stored. (1) Size of copy msg is limited to bound running time. (2) Inefficient handling of missing copy msg. Host Machine Dissemination Data Radio Packets Sensor Node External Flash Previous New For others

19 Optimizing Difference Delivery Solution: Separate difference delivery and decoding. (1) Stores the difference script in the first step (2) Rebuilds the program after receiving decode command. Script Commands Previous New Script Commands Previous New CMD_DOWNLOADING Copy Download Download Copy CMD_DECODE_SCRIPT Copy Download Download Copy download download copy

20 Results Using Rsync with separate decode The performance of using Rsync algorithm with separate decode command is similar to just using Rsync. But, the performance for changing a constant has improved (speed up of 9.1). Speedup Fixed Block Comparison Rsync Rsync + decode Case 1 Modifying Case 2 Comment Case 3 XnpBlink to Case 4 Comment Case5 Comment Constant Lines XnpCount IntToLeds IntToRfm

21 Conclusion Faster network reprogramming using incremental update. Platform independent solution using Rsync algorithm. Speed-up over non-incremental delivery for changing a few lines. 9.1 for changing a constant. Future Work Extension for multi-hop delivery.