Ad hoc Servce Grd A Self-Organzng Infrastructure for Moble Commerce Klaus Herrmann, Kurt Gehs, Gero Mühl Berln Unversty of Technology Emal: klaus.herrmann@acm.org Web: http://www.vs.tu-berln.de/herrmann/ Oslo, Norway, September 17 th 2004 MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 1
Outlne > Ad hoc Servce Grd > General vson, advantages, and challenges > Research Focus > Self-organzng Servce Dstrbuton > Complementng Concepts > Summary and Conclusons MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 2
Wreless Servces at medum-szed Locatons > Locatons: > Constructon stes, hosptals, shoppng malls etc. > Servces (e.g. at a shoppng mall) > Local, faclty-specfc servces for local users > Examples: navgaton, product fnder, reservaton (e.g. restaurant) > Usng cellular phone networks > Non-local communcaton, expensve, low-bandwdth > Usng WLAN access pont technology > Wrng s extremely expensve(!), nflexble, centralzed server MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 3
Ad hoc Servce Grd > Basc dea: Use an ad hoc network > Dstrbuton of PC-lke computers (Servce Cubes) at the locaton > Wreless network nterface, power connector, no perpherals > Drect communcaton between neghborng Servce Cubes > Mult-hop communcaton between Cubes that are further apart > Users access servces va nearest Servce Cube > Advantages > Communcaton s free of charge, modest expenses for setup > No hgh ntal expenses for monolthc central server > Flexbly scalable: addng or removng Cubes durng runtme s easy MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 4
Example Setup: Shoppng Mall 90 m 180 m MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 5
General Challenges > Decentralzaton and Self-Organzaton > Dstrbuted resources Control and organzaton s dffcult > Servce nfrastructure should be nvsble > Mnmal manual nterventons > Self-organze and adapt to changng condtons > Personalzaton vs. prvacy and securty > Offer personalzed servces whle provdng prvacy > Interactons must be secure > Busness Models > Indrect revenue MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 6
Current Research Focus > Self-organzng dynamc servce dstrbuton > Dynamc replcaton and node selecton to meet current demand > Maxmze QoS: response tmes perceved by users > Mnmze network load, balance processng load > Servce lookup and dscovery > Enable users to dscover servces and fnd best servce replca > Data consstency > Acheve data consstency among replcated stateful servces > What does an overall ASG Mddleware/Servceware look lke? MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 7
Self-organzng Servce Dstrbuton > Installaton: one servce replca postoned arbtrarly > Clents start accessng the servce > Assumpton: Spatal dstrbuton of requests s non-unform > General Approach: Use request patterns to gude dstrbuton > Clents always choose closest servce > Request tree T s recorded at each servce replca s Cube > Servce s replcated or mgrated to request hot spots MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 8
Dstrbuton algorthm > Runs perodcally at the replca s Cube > Compute weghtng functon M n for each node n n the tree > Fnd nodes and j n request tree T such that > and j are not n the same subtree > M > M j > M k for all k wth k j > Mgrate servce to node f t s domnatng (M >> M k for k) > Replcate servce to and j f both are domnatng and the servcespecfc replca lmt has not been reached > Dssolve replca f dle for too long MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 9
Weghtnng Functon M n M n = ( D + ) ( ) n 1 Rn t = t k > Informally: Number of transmssons caused by n > Inputs > D n : Hop Dstance of node n from servce s node > R n (): Number of requests transmtted by node n at tme ndex > t: the current tme ndex > k: length of relevant request hstory tme wndow MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 10
Smple Example 1 2 3 4 5 Requests produced: 10 10 10 10 10 R n (): 50 40 30 20 10 D n +1: 1 2 3 4 5 =100 (transmtted) M n : 50 80 90 80 50 1 2 3 4 5 Requests produced: 10 10 10 10 10 R n (): 10 20 50 20 10 D n +1: 3 2 1 2 3 =60 (transmtted) M n : 30 40 50 40 30 MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 11
MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 12 Oscllaton Avodance > Mantan a hstory of adaptatons performed locally at each node > Adaptaton = (Destnaton, Request Tree) > Check for past adaptatons wth smlar Request Trees before performng an adaptaton > Smlarty of two trees T 1 and T 2 s gven by ( ) { } { } wth 1 2 2 1 1 2 1 2 1 2 1 2 1 + = r N r N N N M M M M T T s / /, = k k k k k j T r T N N M root node from ID of node IDs from Set of ff 0 : :
Emergent Effects > Replcas fnd postons where traffc s balanced > None of the nodes nvolved n the request flow stands out n terms of network load produced (no domnatng nodes) > Tunable parameter: Domnaton Factor > Preset lmt on per-servce number of replca controls the average dstance between servce and clents > Tunable Parameter: Replca Lmt > Oscllaton avodance reduces unnecessary adaptatons whle stll keepng the system reactve > Tunable Parameter: Smlarty Threshold > Processng load s balanced > Replcaton and choce of nearest servce by clents MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 13
Result Adaptve Reducton n overall Traffc Overall Transmssons #Transmssons 180 160 140 120 100 80 60 40 20 0 49900 99900 149900 199900 Tme [smulaton steps] MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 14
Work not Covered n the Talk > Dstrbuted lookup servce for moble servces > Forwardng of clent requests to current servce locaton > Lazy propagaton of locaton changes by snoopng meta nformaton pggybacked n servce reples Self-reparng > Data consstency n stateful servces > Weak, optmstc consstency model (nspred by Bayou) > Current work! > Archtectural mplcatons on overall mddleware > Puttng t all together > Past, current, and future work! MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 15
Summary and Conclusons > Ad hoc Servce Grd: Basc vson for a servce provsonng platform for medum-szed locatons > Conceptual groundwork (algorthms and protocols) > Self-organzng servce dstrbuton > Smple, usage-drven algorthm > Transmsson hot spots attract servces untl network load s balanced > Oscllaton s damped whle the system remans reactve to changes > Network load s reduced MOBIS 04 - Oslo - 17.9.4 Herrmann/Gehs/Mühl - Ad hoc Servce Grd 16
Thank you. Queston and comments are welcome. Klaus Herrmann klaus.herrmann@acm.org Intellgent Networks and Management of Dstrbuted Systems Berln Unversty of Technology www.vs.tu-berln.de
Telecommuncatons Insttute Faculty IV Electrcal Engneerng & Computer Scence TU Berln phone: +49 30 314-79830 fax: +49 30 314-24573 offce@vs.tu-berln.de Secretary EN 6 Enstenufer 17 EN-Gebäude D-10587 Berln Germany Intellgent Networks and Management of Dstrbuted Systems