1 Hardware architectures Point-to-point networks Many different - PDF document

Networks in embedded systems Networking for Embedded Systems • Why we use networks. • Network abstractions. • Example networks. initial processing more processing • Properties of wireless networks. PE sensor PE PE actuator PEs may be CPUs or ASICs. 1 2 Why distributed? Network abstractions • Higher performance at lower cost. • The Open Systems Interconnection (OSI) model for networks • Physically distributed activities---time constants may not allow transmission to central site. • Provides a standard way to classify network components and operations. • Improved debugging---use one CPU in network to debug others. • Fault tolerance • May buy subsystems that have embedded processors. 3 4 OSI model OSI layers • Physical: connectors, bit formats, etc. • Data link: error detection and control across a end-use interface application single link (single hop). presentation data format • Network: end-to-end multi-hop data session application dialog control communication. connections transport • Transport: provides connections; may optimize end-to-end service network network resources. reliable data transport data link – TCP, UDP. physical mechanical, electrical 5 6 1

Hardware architectures Point-to-point networks • Many different types of networks: • One source, one or more destinations, no data switching (serial port): – topology; – scheduling of communication; – routing. PE 1 PE 2 PE 3 link 1 link 2 7 8 Bus networks Bus arbitration • Common physical connection: • Priority: Same order of resolution every time. • Fair: every PE has same access over long periods. – round-robin: rotate top priority among PEs. PE 1 PE 2 PE 3 PE 4 fixed A B C A B C round-robin header address data ECC packet format A B C B C A A,B,C A,B,C 9 10 Crossbar Crossbar characteristics • Non-blocking. • Can handle arbitrary multi-cast combinations. • Expensive: Size proportional to n 2 . out4 out3 out2 out1 in1 in2 in3 in4 11 12 2

Multi-stage networks Message-based programming • Use several stages of switching elements. • Transport layer provides message-based programming interface: • Often blocking. send_msg(adrs,data1); • Often smaller than crossbar. • Data must be broken into packets at source, reassembled at destination. • Data-push programming: receivers respond to new data. 13 14 I 2 C bus I 2 C physical layer • Designed for low-cost, medium data rate applications. • Characteristics: master 1 master 2 – serial; data line SDL – multiple-master; clock line SCL – fixed-priority arbitration. • Several microcontrollers come with built-in I 2 C slave 1 slave 2 controllers. 15 16 I 2 C signaling I 2 C data link layer • Sender pulls down bus for 0. • Every device has an address (7 bits in standard, 10 bits in extension). • Sender listens to bus---if it tried to send a 1 and – Bit 8 of address signals read or write. heard a 0, someone else is simultaneously transmitting. • General call address allows broadcast. • Transmissions occur in 8-bit bytes. 17 18 3

I 2 C bus arbitration I 2 C transmissions • Sender listens while sending address. • When sender hears a conflict, if its address is multi-byte write higher, it stops signaling. S adrs 0 data data P • Low-priority senders relinquish control early read from slave enough in clock cycle to allow bit to be transmitted reliably. S adrs 1 data P write, then read S adrs 0 data S adrs 1 data P 19 20 Ethernet Ethernet topology • Dominant non-telephone LAN. • Bus-based system, several possible physical layers: • Versions: 10 Mb/s, 100 Mb/s, 1 Gb/s 10 Gb/s. • Goal: reliable communication over an unreliable medium. A B C 21 22 Networking for Embedded CSMA/CD Systems • Carrier sense multiple access with collision • Why we use networks. detection: • Network abstractions. – sense collisions; • Example networks. – exponentially back off in time; • Properties of wireless networks. – retransmit. • No prioritization • Can’t guarantee real-time deadlines. However, may provide good service at proper load levels. 23 24 4

RF Communication In Real World Environments • Notoriously unpredictable – Variable environment noise – Non-linear signal strength decay – Multi-path effect – Transmission collision • Especially in a wireless sensor network • In-door – Low-power low-cost radio – ISI office – Large-scale high-density deployment Habitat • – Harsh environment – Topanga state park • Unobstructed The following slides are borrowed from Jerry Zhao’s SenSys’03 presentation – ISI parking lot 25 26 TinyOS Network Stack (MICA) New Functionalities Application Application ~60 MICA Motes ARQ – RFM Radio ReTx ACK Rx 433Mhz, TinyOS MAC: CSMA TinyOS MAC: CSMA – TinyOS 1.0 SECDED SECDED Manchester 4B6B RFM Transceiver RFM Transceiver 27 28 Setup Packet Delivery Performance • Packet Delivery Performance on Physical Layer • Spatial and temporal characteristics of packet loss (Single-hop) • Its dependence on environment – Node placed on a line with single transmitter • Its impact on sensor network design – Fine-grain measurement of each metric – Environmental impact • Packet Delivery Performance on MAC Layer (Multi-hop) – “Realistic” deployment with artificial traffic load – Transmission collision – ARQ recovery 29 30 5

Spatial Profile of Packet Delivery Gray Area in Packet Loss • Relatively large region of poor connectivity In-door – Across a wide variety of • Node positions environments 4B6B Encoding – Spanning as large as 30% of High Tx Power the effective transmission In-door range 2hrs (7200 pkts) Unobstructed Habitat Out-door “Gray Area” is evident in the communication range 31 32 Reception Rate Changes Over Time Standard Deviation in Packet Loss Window size = 40 Window size = 40 Pkts 4B6B Encoding 4B6B Encoding High Tx Power High Tx Power In-door Each data point is from 180 samples Variability over time with large dynamic range 33 34 Can We Filter Bad Links From Impact on System Design Received Signal Strength Reading? • Assume circular RF coverage: 30% in Radius => 51% in Neighborhood! • Need to continuously monitor 4B6B Encoding link quality to filter out High Tx Power unreliable neighbors In-door Cannot use a simple threshold 35 36 6

Can We Use Stronger Encoding? Possible Reasons for Gray Area • Multi-path effects • Environment noise • Difference between individual transceivers 4B6B 1byte=>1.5 bytes SECDED 1byte=>3byte High Tx Power In Door Gray area can be masked (not eliminated) but with high overhead 37 38 Asymmetric Links (MAC Layer) Summary • Properties of RF links in real world – Heavy tail of packet loss distribution Metric: – Evident gray area in communication range Packet loss difference – High variance over time – Incapability of simple ARQ 4B6B Encoding High Tx Power • The methodology and solutions are helpful In Door for other systematic measurements. Common existence of asymmetric links 39 40 Acknowledgements • Slides 1-25 are modified from Wayne Wolf’s textbook slides • Slides 25-41 are modified from Jerry Zhao’s SenSys’03 slides 41 7