330 likes | 342 Views
This overview discusses wireless connectivity, end-host mobility, and ad-hoc routing problems, along with various solutions such as modifying transport and link layer protocols, explicit loss signal, retransmission, forward error correction, and packet shrinking.
E N D
CS 268: End-Host Mobility and Ad-Hoc Routing Ion Stoica Feb 18, 2004
Overview • Wireless • End-host mobility • Ad-hoc routing
Wireless • Wireless connectivity proliferating • Satellite, line-of-sight microwave, line-of-sight laser, cellular data (CDMA, GPRS, 3G), wireless LAN (802.11a/b), Bluetooth • More cell phones than currently allocated IP addresses • Wireless non-congestion related loss • Signal fading: distance, buildings, rain, lightning, microwave ovens, etc. • Non-congestion related loss • Reduced efficiency for transport protocols that depend on loss as implicit congestion signal (e.g. TCP)
Problem Best possible TCP with no errors (1.30 Mbps) TCP Reno (280 Kbps) Sequence number (bytes) Time (s) 2 MB wide-area TCP transfer over 2 Mbps Lucent WaveLAN (from Hari Balakrishnan)
Solutions • Modify transport layer • Modify link layer protocol • Hybrid
Modify Transport Protocol • Explicit Loss Signal • Distinguish non-congestion losses • Explicit Loss Notification (ELN) [BK98] • If packet lost due to interference, set header bit • Only needs to be deployed at wireless router • Need to modify end hosts • How to determine loss cause? • What if ELN gets lost?
Modify Link Layer • Advantages: • Limited changes: only link-layer affected • Preserve end-to-end (TCP) semantics • Three types of losses • Total packet loss • Partial packet loss • Packet corrupted by bit errors • Three methods to reduce packet loss • Packet retransmission • Forward error correction • Packet shrinking
Retransmission • Advantages: • Optimal overhead: only lost packets are retransmitted • Disadvantages: “nasty” interactions between TCP control look and link-level retransmission • Both TCP and link-layer can retransmit same packets • Can introduce packet reordering • Can introduce highly variable delays
FEC • Forward Error Correction (FEC) codes • k data blocks, use code to generate n>k coded blocks • Can recover original k blocks from any k of the n blocks • n-k blocks of overhead • Trade bandwidth for loss • Can recover from loss in time independent of link RTT • Useful for links that have long RTT (e.g. satellite) • Pay n-k overhead whether loss or not • Need to adapt n, k depending on current channel conditions
FEC & Packet Shrinking • Advantages: • No changes at end hosts or base-stations above link-layer • Decrease packet loss • Do not introduce variability • Disadvantages: • Overhead can be quite high, e.g., packet segmentation/reassembly, encoding/decoding
Flex [Eckhardt &Steenkiste ’98] • Combine the three types of error control seven policies (three fixed and four adaptive) • Most sophisticated : Flex • When two or more packets in a window of ten are truncated reduces “safe” packet size by 85% • When three consecutive packets do not experience truncation linearly increase packet size • When two or more packets in a window of ten cannot be decoded decrease user data by 15% (more conservative coding) • When three consecutive packets can be decoded increase user data linearly • Note: adaptation exhibits a linear-increase multiplicative-decrease behavior
Hybrid: Indirect-TCP [Bakre & Badrinath ’94] • Split TCP connection into 2 TCPs • Advantages • Optimize performance for wireless TCP • No changes to protocol for fixed hosts (transparent to fixed hosts) • Disadvantages • Violate end-to-end TCP semantics (why?) • High overhead, because dual stack at BS • Might introduce high delays because packet buffering wireless TCP regular TCP Internet Mobile Host (MH) Fixed Host (FH) Base Station (BS)
Hybrid: Snoop-TCP [Balakrishnan et al. ’95] • Insert a “snoop agent” between fixed host (FH) and mobile host (MH) • Monitor traffic, retransmit packets and discard acknowledgements • Notes: • Avoid violating end-to-end semantics • What about layering? TCP Internet packet retransmissions Mobile Host (MH) Fixed Host (FH) Base Station (BS)
Overview • Wireless • End-host mobility • Ad-hoc routing
Motivation and Problem • Network Layer mobility • Movement = IP address change • Problem: • Location • I take my cell phone to London • How do people reach me? • Migration • I walk between base stations while talking on my cell phone • I download or web surf while riding in car or public transit • How to maintain flow?
Solutions • Mobile IP (v4 and v6) • TCP Migrate • Multicast
Mobile IP • Use indirection to deal with location and migration • Point of indirection: Home Agent (HA) • Resides in Mobile Host’s (MH) home network • Uses MH’s home IP address • As MH moves, it sends its current IP address to HA • Correspondent Host (CH) contacts MH through HA • HA tunnels packets to MH using encapsulation • MH sends packets back to CH • Tunnels packets back to HA (bi-directional tunneling) • Sends directly to CH (triangle routing)
Mobile IP Properties • Advantages • Preserves location privacy • CH does not have to be modified • Disadvantages • Triangle routing and especially bidirectional tunneling increase latency and consume bandwidth • HA is single point of failure
Mobile IP Route Optimization • CH uses HA to contact MH initially • MH sends its location directly back to CH • CH and MH communicate directly • Lose location privacy • CH must be modified
TCP Migrate [SB00] • Location: uses dynamic DNS updates • When MH moves to new IP address, it updates its home DNS server with new hostname to IP address mapping • Migration: • When MH moves, it sends update to CH • Advantage • No new infrastructure • Incremental deployable • Efficient routing • Disadvantages • Only works for TCP • Both CH and MH need new TCP implementation • No location privacy
Receiver (R1) id id R1 R2 Receiver (R2) i3 Based Mobility (Z+03) • Receiver R maintains a trigger (id, R) in the i3 infrastructure; sender sends packets to id • Advantages • Support simultaneous mobility • Efficient routing: receiver can chose id to map on a close i3 server • Ensure privacy • Disadvantage • Require a new infrastructure Sender
Other solutions • Network specific mobility schemes • Cellular phones, 802.11b • Cannot handle mobility across networks (e.g. move laptop from cell phone to 802.11b) or between same network type in different domains (e.g. laptop from Soda Hall 802.11b to campus 802.11b) • Other mobility models • Terminal/personal mobility: • e.g.accessing email through IMAP from different computers • Session mobility: • e.g. talking on cell phone, transfer call in progress to office phone
Summary • Not that important today • Few portable, wireless IP telephony devices • Cell phones have their own network-specific mobility schemes • IP-based wireless networks are not ubiquitous enough to be seamless • PDA (e.g. palm pilot) are too weak to do handle long-lived flows • Future • Cellular networks will become IP-based, need IP mobility scheme • PDA are becoming more powerful
Overview • Wireless • End-host mobility • Ad-hoc routing
Motivation • Internet goal: decentralized control • Someone still has to deploy routers and set routes • Ad Hoc routing • Every node is a router • Better wireless coverage • Better fault tolerance (e.g. node bombed, stepped on, exhausted power) • No configuration (e.g. temporary association) • Dedicated router costs money
Routing • DSDV: destination-sequenced distance vector • TORA: Temporally-Ordered Routing Algorithm • DSR: Dynamic Source Routing • AODV: Ad hoc On-demand Distance Vector • TORA, DSR, and AODV are all on-demand routing protocols
DSDV • Hop-by-hop distance vector • Routing table contains entries for every other reachable node • Nodes pass their routing tables to neighbors periodically • Routing tables are updates using standard distance vector algorithm • Old routes are ignored using sequence numbers • O(n) routing state / node, O(n*k) communication size / node / period • k = average node degree
TORA • Temporally-Ordered Routing Algorithm • Interested in finding multiple routes from SD • Find routes on demand • Flood query to find destination • Flood query response to form multiple routes • O(m) routing state / node, O(n*k) communication / node / route update • m = nodes to communicate with; worst case m=n
DSR • Dynamic Source Routing • Packet headers contain entire route • Flood query to find destination • Intermediate nodes don’t have to maintain routing state • Nodes listen for and cache queries, responses as optimization • Nodes gratuitously sends response packets to shorten paths when they hear packets with sub-optimal routes • O(m) routing state / nodes, O(n*k) communication / node / route update • Smaller constant than other protocols • O(n1/k) space required in header
AODV • Ad Hoc On-Demand Distance Vector • Flood query to find destination • Reply is sent back to source along the reverse path • Intermediate nodes listen for reply to set up routing state • State is refreshed periodically • O(m) routing state / node, O(n*k) communication / node / route update
Results • Avoid synchronization in timers • TORA does not scale to 50 node • Suffers control traffic congestion collapse • DSDV fails to deliver packets when movement is frequent • Only maintains one route/destination • AODV has high routing overhead when movement is frequent • Combination of DSDV maintenance of state + flooding of DSR • DSR does well compared to others • Designed by authors not surprising! • [LJC+00] shows congestion collapse beyond 300 nodes
Related Work • Greedy Perimeter Stateless Routing (GPSR) [Karp and Kung, Mobicom 2000] • Separate addressing from naming • Assume everyone has GPS • Do Cartesian routing • Separate scalable, efficient, fault tolerant service to map from names to addresses • How to deal with selfish users? [MGL+00] • Listen to neighbors to make sure they are forwarding • Convey black list information back to source • Route around selfish nodes
Conclusions • Proliferation of wireless network interfaces provide ready market • Ad hoc provides less configuration, more fault tolerance, better coverage, lower cost • Many interesting and unsolved problems