H AZY S IGHTED L INK S TATE Eleonora Borgia IIT – CNR Pisa - Dicember 4th, 2003

Link State Routing Link State Routing is the most widely used approach in Internet; but it scales poorly in mobile Ad Hoc networks. In literature there are many approaches with the goal to reduce the overhead introduced by it: Efficient Dissemination : updates sent throughout the network more efficiently (e.g. OLSR, TBRF, STAR) Limited Dissemination : restriction of the scope of routing update in space and time (e.g. hierarchically link state, FSR, GSR)

TRADITIONALLY: OVERHEAD  CONTROL OVERHEAD Amount of bandwidth require to construct and maintain routes: PROACTIVE: number of packets exchanged between nodes in order to maintain node’s routing table REACTIVE: consumed bandwidth for Request/Reply messages As N increases, keeping route optimality imposes an unacceptable cost and a proper performance evaluation based only on those types of overheads is not correct. Total overhead (1) The impact of SUB-OPTIMAL routes and its overhead must be taken into account.

Ex: A = 3 hops B = k hops SUB-OPT ov = (k-3)*Packet_length Total overhead (2) NEW DEFINITION: TOTAL OVERHEAD of X is the sum of: 1.PROACTIVE OVERHEAD : Amount of bandwidth consumed by X in order to propagate routing information BEFORE it is needed (periodically and/or event- driven) 2.REACTIVE OVERHEAD : Amount of bandwidth consumed by X to build paths S/D on-demand (AFTER S generates a traffic flow to D) 3.SUB-OPTIMAL OVERHEAD : Difference between the bandwidth consumed using the sub-optimal paths and that eventually consumed if the data had followed the shortest available path(s) SD A B

Achievable region of overhead ACHIEVABLE REGION: Three dimensional region formed by all overheads (proactive, reactive, sub-optimal) can be induced by any protocol under the same scenario (traffic, mobility, etc.) In a bidimensional graph: Convex region, lower-bounded by the curve of overhead points achieved by the “efficient” protocols (i.e. minimazing some source of overhead given a condition imposed on the others) The best operating region is where all overheads are present and balanced

FUZZY SIGHTED LINK STATE family FSLS Family is based on the observation that nodes that are far away not need to have a complete topological information in order to make a good next hop decision Proactive Protocols family with Limited Dissemination : Every t i sec an update message is sent in a region of S i hops; As S i increases, update’s frequencies decrease Many approaches can be implemented varing the values (t i, S i ). Choose {S i } sequence that minimizes the total overhead of the FSLS family TARGET:

HAZY SIGHTED LINK STATE (HSLS) 1.Every t b sec a global LSP (TTL=  ) is sent in the entire network to give a complete overview of the network topology; 2.After one global LSP, a node wakes up every t e sec (t e <t b ) and sends a LSP with TTL=2 if there has been a link status change in the last t e sec; 3.Every 2 i-1 * t e sec (with i=1,2,3..) a node wakes up and sends a LSP with TTL= 2 i if there has been a link status change in the last 2 i-1 * t e sec. S i = 2 i

Comparative study Routing Protocol Total_overheadCasesSCALABILITY PF  ( t * N 2 ) Always Scalable w.r.t t & lc No Scalable w.r.t N SLS  ( lc * N 2 ) Always Scalable w.r.t t No Scalable w.r.t N & lc DSR  ( s *N 2 + t *N 2 *log 2 N) No route cache Scalable w.r.t t & lc No Scalable w.r.t N HierLS  ( lc * N 1,5 + t *N 1,5+  ) LM proactive Almost Scalable w.r.t N Scalable w.r.t t No Scalable w.r.t lc ZRP  ( lc * N 2 )  ( lc 1/3 * s 2/3 * N 5/3 )  ( s * N 2 ) if lc = O ( lc /  N) if lc =  ( lc /  N) and lc = O ( s *N) if lc =  ( s *N) Scalable w.r.t t No Scalable w.r.t N & lc HSLS  ((  lc t ) * N 1.5 )  ( lc * N 1,5 ) if lc = O ( t ) if lc =  ( t ) Scalable w.r.t t & N No Scalable w.r.t lc