Zigbee Network Performance

Transcription

1 Wireless Control That Simply Works Zigbee Network Performance Typical Results and Implications for Application Design Skip Ashton, Ember Corporation Copyright 2004 ZigBee TM Alliance. All Rights Reserved.

2 Testing Goals Establish Baseline of Zigbee network performance Join times may or may not be important based on application Throughput is an often requested parameter for application design Latency per hop and for the system Maximum messaging rate the network can sustain These results are initial test results and only represent one software implementation Evaluate single chip, network coprocessor and two chip solutions This type of testing will need to include multi-vendors to ensure interoperability does not effect results 2

3 Test Methodology Join Test Configuration Throughput Test Configuration Testing Done Using: EM2420/Atmega 128 and EM250 Standard development hardware kits 31 devices in join testing and up to 7 hops in throughput and latency EmberZNet Software all using table routing not tree Time stamping of packets using debug backchannel and packet trace Varied message type, packet size, security Test application written to initiate tests and gather output Nodes are not specially configured for these tests but are operating under expected typical configuration 3

4 Join Time Join Test Results Use of security on joining results in 150 millisecond longer join time Number of parents has impact on success rate due to broadcast collisions Join times need to be considered for new device joining network or mobile/sleeping device that has lost parent and is reassociating Impact of Security % Succeed Avg Join Time St Dev No security EM Security Level 5, unsecure joining EM Security Level 5, secure joining EM Impact of Number of Parents One Parent Many Parents, One permit joining Many Parents, All permit joining

5 Expected Throughput Calculated theoretical payload throughput MAC only (102 byte payload) kbps APS no security (88 byte payload) kbps APS w/security (70 byte payload) 79.5 kbps This is based on the MAC delays and Zigbee packet headers. Node processing delays will result in lower throughput than these results Multihop network also reduces throughput 5

6 Calculated Impact of Node Processing Delay Calculated Maximum Throughput This represents calculated single hop performance only. Variation is due to payload capacity Applicaiton Throughput MAC APS APS w/security Node Processing Delay (milliseconds) 6

7 Typical Throughput APS Messages Zigbee APS Messages EM250 - No Security, No Retry Application Throughput Throughput Data is for 91 byte payload Highest throughput at single hop smallest interpacket delay Peaks at 46 kbps for application throughput Performance drops after 2 hops due to packet loss Even at 5 hops, performance is higher than 25 kbps S1 Hops S3 S millisecond delay Note: Throughput is based on expected throughput given the interpacket spacing and adjusted based on percent of successful packets from the test 7

8 Typical Throughput Adding APS Reply Zigbee APS MessageEM250 - No Security, APS retry Throughput Data is for 91 byte payload Highest throughput at single hop with smallest interpacket delay Peak remains at 46 kbps for application throughput A pplication thoug h p ut Performance drops quickly as reply consumes additional bandwidth There is a throughput penalty for knowing if message was delivered S1 hops S3 S millisecond delay

9 Typical Throughput Adding Security S1 hops S3 S5 S7 Zigbee APS Messages EM250 - Security, No retry Throughput Data is for 73 byte payload (reduced maximum payload due to security) Highest throughput at single hop smallest interpacket delay Peaks at 37 kbps for application throughput millisecond delay A pplic a tio n th ro u g h p ut Smaller max payload decreases maximum throughput Performance drops after 2 hops due to packet loss Even at 7 hops, performance is higher than 15 kbps 9

10 End to End Latency End to End Latency - EM250 APS Message Latency Data is for 5 and 73 byte payload Data is measured at application interface milliseconds byte Sec 73 byte Sec 5 byte -No Sec 73 byte No Sec Increases linearly per hop count Packet size and security increases per hop latency Hops 10

11 Per Hop Latency Per Hop Latency EM250 APS Messages Latency Data is for 5 and 73 byte payload Previous data is divided by hop count milliseconds per hop Illustrates a clear per hop latency that varies with packet size and security Hops 5 byte Sec 73 byte Sec 5 byte -No Sec 73 byte No Sec 11

12 Round Trip Latency Round Trip Latency - EM250 APS Messages Latency Data is for 5 and 73 byte payload Measured at application interface Increases linearly per hop count Milliseconds byte - Sec 73 byte - Sec 5 byte - No Sec 73 byte - No Sec Packet size and security increases per hop latency All data is higher than one way messages, but not double Hops 12

13 Round Trip Latency Per Hop Round Trip Latency Per Hop Latency - EM250 APS Reply 5 byte Sec 73 byte Sec 5 byte -No Sec 73 byte Sec Latency Data is for 5 and 73 byte payload Previous Data is divided by hop count The per hop latency that varies with packet size and security does not double with round trip message Hops 13

14 Transmit Time of Packet Removed Per Hop Latency - EM250 APS Messages OTA Time Removed Removal of OTA time results in 5 and 73 byte packets without security having similar latency Security impacted by packet size due to time required for processing milliseconds Hops 5 byte -No Sec 73 byte No Sec 5 byte Sec 73 byte Sec 14

15 Impact of Two Chip Solution Zigbee APS Messages EM Security, No retries This is using EM2420 and Atmega 128 processor. Security processing done using AES engine in EM2420 in stand alone mode Application throughput Throughput Data is for 73 byte payload Highest at single hop smallest interpacket delay Peaks at 26 kbps instead of 37 kbps for single chip hops S1 S2 S3 S4 S S millisecond delay 0 Performance drops rapidly to about 10 kbps Use of two chips decreases steady state expected throughput from 15 kbps to about 10 kbps when using security 15

16 Two Chip Latency End to End Latency - EM2420 APS Messages Latency Data is for 5 and 73 byte payload Increases linearly per hop count similar to single chip m illiseconds Security increases per hop latency even higher than on single chip Hops 5 byte No Sec 73 byte - No Sec 5 byte - Sec 73 byte - Sec 16

17 Two Chip Per Hop Latency Per hop Latency Per Hop Latency - EM2420 APS Messages Latency Data is for 5 and 73 byte payload Increases linearly per hop count similar to single chip Per hop latency is higher than single chip Security impact is larger Hops 5 byte No Sec 73 byte - No Sec 5 byte - Sec 73 byte - Sec 17

18 Two Chip Time with Transmit Time Removed Per Hop Latency EM2420 APS Message Latency - OTA Time Removed Results for 5 and 73 byte similar and close to single chip with security off. Impact of security is substantially higher in two chip solution. m illiseconds hops 5 byte No Sec 73 byte - No Sec 5 byte - Sec 73 byte - Sec 18

19 Application Design Application Data models must reflect expected throughput Join time for mobile nodes must be considered in application design Throughput rate needs to be considered when transferring fixed blocks of data Exceeding bandwidth results in lost messages normally application would need to backoff in this case Use of reply provides indication message was received, but lowers overall throughput Per hop latency is consistent but varies with packet size and security settings application can set packet size and security settings but expected hop count depends on the topology Use of security lowers overall throughput and increases latency Application designers should turn on security early in product testing to ensure impact on system is understood Use of Single Chip solutions reduces the impact of security on throughput and latency 19

20 Additional Work Needed Analysis and optimization of results and software performance This represents some of the early performance data and some areas of improvement are expected as results are analyzed Improvements expected in stack processing time and security processing time Many to one routing throughput Gateway device represents bottleneck and needs to be characterized to allow system design to avoid this bottleneck Network processor testing needs to be completed Multi-vendor testing to evaluate the impact Changes to Zigbee specification and impact on results New mobile node procedure expected to be quicker than these results Windowing method in Zigbee fragmentation should improve throughput Reductions in payload will have incremental impact on application throughput 20

21 Thank You! Skip Ashton