Approximate Semantics for Wirelessly Networked Applications - PowerPoint PPT Presentation

Approximate Semantics for Wirelessly Networked Applications ☞ Benjamin Ransford ☜ Adrian Sampson Luis Ceze WACAS ’14 // March 2, 2014 � 1

Algorithms ✅ Programming Languages ✅ Compilers ✅ Storage ✅ Arithmetic Units ✅ WACAS ’14 // March 2, 2014 � 2

Algorithms ✅ Programming Languages ✅ Compilers ✅ Storage ✅ Communication ❌ Arithmetic Units ✅ WACAS ’14 // March 2, 2014 � 2

WACAS ’14 // March 2, 2014 � 3

foo.bmp WACAS ’14 // March 2, 2014 � 4

foo.bmp ✅ ✅ ✅ ✅ WACAS ’14 // March 2, 2014 � 4

foo.bmp ✅ ✅ ✅ ✅ WACAS ’14 // March 2, 2014 � 4

foo.bmp ✅ ✅ ✅ ✅ WACAS ’14 // March 2, 2014 � 4

foo.bmp ✅ ✅ ❌ ✅ ❌ ✅ ✅ ✅ WACAS ’14 // March 2, 2014 � 4

foo.bmp ✅ ✅ ❌ ✅ ❌ ✅ ✅ ✅ WACAS ’14 // March 2, 2014 � 4

● 40 ● Frames Retransmitted (%) ● 30 20 ● ● 10 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 ● ● ● ● ● ● 6 12 18 24 30 36 42 48 54 WiFi Bitrate (Mbps) 12m apart @ UW CSE WACAS ’14 // March 2, 2014 � 5

802.11 frame header 802.11 frame payload IP header IP header checksum 802.11 frame IP payload checksum TCP header TCP checksum Application data WACAS ’14 // March 2, 2014 � 6

● 40 ● Frames Retransmitted (%) ● 30 20 ● ● 10 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 ● ● ● ● ● ● 6 12 18 24 30 36 42 48 54 WiFi Bitrate (Mbps) 12m apart @ UW CSE WACAS ’14 // March 2, 2014 � 7

● 40 ● Frames Retransmitted (%) ● 30 < 3% of bits 20 are bad! ● ● 10 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 ● ● ● ● ● ● 6 12 18 24 30 36 42 48 54 WiFi Bitrate (Mbps) 12m apart @ UW CSE WACAS ’14 // March 2, 2014 � 7

S elective A pproximate P rotocol • Optional, partial integrity checks • Suitably generic (can work with “your” apps) • Backward compatible with existing networks • Simple API “SAP” WACAS ’14 // March 2, 2014 � 8

S elective A pproximate P rotocol Move error checking to the application layer! 802.11 frame header 802.11 frame payload IP header IP header checksum 802.11 frame IP payload checksum TCP header TCP checksum Application data WACAS ’14 // March 2, 2014 � 9

S elective A pproximate P rotocol Move error checking to the application layer! 802.11 frame header 802.11 frame payload IP header IP header checksum 802.11 frame IP payload checksum TCP header TCP checksum Application data WACAS ’14 // March 2, 2014 � 9

S elective A pproximate P rotocol Move error checking to the application layer! 802.11 frame header 802.11 frame payload IP header IP header checksum 802.11 frame IP payload checksum TCP header TCP checksum Application data Application TCP 802.11 WACAS ’14 // March 2, 2014 � 9

S elective A pproximate P rotocol Move error checking to the application layer! 802.11 frame header 802.11 frame payload IP header IP header checksum 802.11 frame IP payload checksum TCP header TCP checksum Application data Application Application TCP UDP-Lite 802.11 802.11 WACAS ’14 // March 2, 2014 � 9

Applications WACAS ’14 // March 2, 2014 � 10

Open Questions • Encryption, compression, encoding challenges � • How best to integrate quality metrics? � • How to expose quality controls to apps? WACAS ’14 // March 2, 2014 � 11

Summary • Mismatch between approximate computing and precise communication � • Relax lower-layer integrity checks � • Improve throughput & range ransford@cs.washington.edu WACAS ’14 // March 2, 2014 � 12