ΗΥ537: Έλεγχος Πόρων και Επίδοση σε Ευρυζωνικά Δίκτυα Βασίλειος Σύρης Τμήμα Επιστήμης Υπολογιστών Πανεπιστήμιο Κρήτης Εαρινό εξάμηνο 2008 Διαδικαστικά Περιεχόμενο και στόχος μαθήματος Έλεγχος δικτύων - Βασικοί μηχανισμοί - Προβλήματα
HY537: Έλεγχος Πόρων και Επίδοση σε Ευρυζωνικά Δίκτυα Διδάσκων: Βασίλειος Σύρης (vsrs@csduocgr) Προαπαιτούμενα: HY335 Δίκτυα Υπολογιστών, ΗΥ217 Πιθανότητες Διδακτικές μονάδες 4, Θεματική περιοχή: Β Βοηθός: Χαρίτον Μελισσάρης Μαθήματα: Δευ 3-5, Τετ 3-5, B211 Σελίδα: http://wwwcsduocgr/~hy537 Πρόγραμμα διαλέξεων, υλικό σε ηλεκτρονική μορφή (διαφάνειες, αναφορές, κτλ) Ηλεκτρονική λίστα: hy537-lst@csduocgr Intro - 3 HY537: Έλεγχος Πόρων και Επίδοση σε Ευρυζωνικά Δίκτυα Διαλέξεις: ~ 10 εβδομάδες (~20 διαλέξεις), υπόλοιπες για παρουσιάσεις εργασιών Ασκήσεις: 4 σειρές Εργασία: Υλοποίηση & μελέτη μηχανισμού ελέγχου Θα δοθούν υποψήφια θέματα: μηχανισμοί ελέγχου σε ενσύρματα και ασύρματα δίκτυα Υλοποίηση του μηχανισμού ελέγχου σε προσομοιωτή (NS-2, MATLAB,Mathematca) Bήματα: Παρουσίαση προβλήματος (στην τάξη) Υλοποίηση και εκτέλεση πειραμάτων Τελική παρουσίαση (στην τάξη) Αναφορά Συζήτηση στην τάξη Τελική εξέταση Intro - 4
HY537: Έλεγχος Πόρων και Επίδοση σε Ευρυζωνικά Δίκτυα Βαθμός: 10% ασκήσεις 50% εργασία 20% δύο παρουσιάσεις & συμμετοχή σε συζητήσεις στην τάξη 30% αναφορά 40% τελική εξέταση (θα πρέπει βαθμός >= 4) Intro - 5 Σκοπός Εις βάθος μελέτη θεμάτων σχετικά με δικτυακές υπηρεσίες, και τις τεχνολογίες για παροχή αυτών Έλεγχος πόρων σε ενσύρματα (IP, ATM, IP-QoS) και ασύρματα (CDMA, Wreless LAN) Κατανόηση μαθηματικών μοντέλων & εργαλείων για μελέτη της απόδοσης δικτύων Εφαρμογή οικονομικών μοντέλων για έλεγχο συμφόρησης και πόρων Ο ρόλος της χρέωσης ως μηχανισμός ελέγχου δικτύων Intro - 6
Περιεχόμενα Βασικές αρχές ελέγχου δικτύων Δικτυακές υπηρεσίες & δικτυακές τεχνολογίες (IP, ATM, IntServ, DffServ, MPLS) Περιορισμοί στα δίκτυα υπολογιστών ισοδύναμο εύρος ζώνης (effectve bandwdth) Οικονομικά μοντέλα και εφαρμογή τους για έλεγχο δικτύων Χρέωση υπηρεσιών εγγυημένης ποιότητας υπηρεσιών (guaranteed Qualty of Servce QoS) Συμφόρηση: TCP, ECN, κα και εξέλιξη μηχανισμών ελέγχου συμφόρησης, διαμοίραση πόρων, δικαιοσύνη Έλεγχος πόρων σε ασύρματα δίκτυα: CDMA, Wdeband CDMA (3G), IEEE 80211 (Wreless LAN WF), IEEE 80216 (WMAX) Intro - 7 Υλικό - αναφορές Prcng for Communcatons Networks: technology, economcs and modellng Costas Courcoubets and Rchard Weber, Wley, Aprl 2003 Hgh-Performance Communcaton Networks, Second Edton Jean Walrand and Pravn Varaya, Morgan Kaufmann, 2000 Ερευνητικά άρθρα από περιοδικά και συνέδρια Intro - 8
New advances and challenges Technologcal advances Deregulaton of telecommuncatons market New hghly compettve envronment Both technology and economcs play major role Success note determned by technology alone Tradtonally engneers devsed servces wthout reference to how they should be prced Intro - 9 Why prcng communcatons servces s exctng Prcng affects ways servces are used Communcaton contracts provde substantal flexblty, can be used for provdng ncentves Advances n network technology provde new possbltes for (economc) sgnalng on fast tme scales Competton nfluenced by network archtecture Communcaton servces are goods that should be prced accordng to qualty and performance Intro - 10
Role of economcs Decentralzed control mechansms Use prce and congeston sgnals to provde ncentves Engneerng performance: n terms of utlzaton, delay, blockng, etc Economc effcency: nclude the value that customers obtan from usng the network Enttes (users-customer, network) are ratonal, seekng to maxmze there own beneft Intro - 11 Network control for varous servce types ATM and IP supports statstcal multplexng Network control dfferent for guaranteed and elastc servces Guaranteed servces User-network contract Call Admsson Control - CAC Open loop control Elastc servces (TCP, ABR): no CAC, except for MCR (Mnmum Cell Rate) Closed loop control Intro - 12
Tme scales of network control Traffc & Congeston Control Functons Selectve cell dscard, frame dscard, prorty control and schedulng, Usage Parameter Control (UPC), traffc shapng Feedback controls Response tme Cell tme Round-trp propogaton tme Routng, Call Admsson Control (CAC) Network management control Connecton nterarrval tme order of mnutes Prcng months, years Intro - 13 Routng and CAC Traffc Contract destnaton source X Routng: fnd path from source to destnaton that fulfls user requrements (bandwdth, QoS) Call Admsson Control (CAC): performed at every swtch, determnes whether there are enough resources to accept a call Intro - 14
Some Network Management Questons User stes SLAs User stes Swtch Network provder Swtch Spare capacty of the lnk, whle guaranteeng a partcular QoS (maxmum delay or loss)? Effect on QoS f one or more users are added? How effectve s traffc shapng? Selecton of traffc parameters of SLA (Servce Level Agreement? Intro - 15 Possble Soluton Approaches Use average load over large ntervals (~mnutes) can lead to volaton of QoS Use worst case scenaros (peak rate) can lead to under-utlsaton of resources Use traffc models real world traffc not well approxmated by traffc models Use actual traffc measurements and advanced statstcal analyss procedures Intro - 16
Smply Measurng Not Enough Current technology can support traffc montorng at very hgh speeds (> 10 Gbts/s) But, a huge amount of data s collected What part of these measurements s mportant? How can they be used for network management? Intro - 17 Example: Resource Usage SLAs a) b) c) Swtch Network provder C Swtch All streams have same duraton and volume What s the resource usage for each stream? Answer: Depends! Small lnk capacty C: a<b,c Medum lnk capacty C: a=b<c Large lnk capacty C: a=b=c Intro - 18
Example: Resource Usage SLAs a) b) c) Swtch Network provder C Swtch All streams have same duraton and volume What s the resource usage for each stream? Answer: Depends! Small lnk capacty C: a<b,c Medum lnk capacty C: a=b<c Large lnk capacty C: a=b=c Intro - 19 Loadng an Elevator wth Boxes w, v What s the relatve effectve usage of a box? Equvalently, n what sense = k x or α 1 =k x α 2 W max, V max w 1, v 1 w 2, v 2 Key noton: substtuton Intro - 20
Loadng an Elevator (cont) What s the relatve effectve usage of a box? Depends on whch constrant s actve: max weght or max volume Determned by operatng pont If max weght s actve, then effectve usage equals box s weght w = W max 2w 1w v < V max Wmax, V max Effectve bandwdth = weght Intro - 21 Loadng an Elevator (cont) If max volume s actve, then effectve usage equals box s volume v = V max w < W max 2v 1v Effectve bandwdth = volume W max, V max Intro - 22
Loadng an Elevator (cont) What s the relatve effectve usage of a box? Depends on whch constrant s actve: max weght or max volume Determned by operatng pont If max weght s actve, then effectve usage equals box s weght w = W 2w max 2w 1w 1w v < V max W max, V max Effectve bandwdth = weght = volume Intro - 23 Example: Resource Usage SLAs a) b) c) Swtch Network provder C Swtch Identfcaton of mportant tme scales ndcate what to measure ndvdual packet traces can be overly detaled aggregate load n ntervals wth duraton a few 10s mllseconds can be suffcent tme scales depend on network resources and traffc mx Intro - 24
Quantfyng resource usage Resource usage depends on technology Crcut swtched networks: crcut wth fxed capacty reserved for whole duraton of connecton Packet swtched networks: no statc bandwdth reservaton Resource usage depends on guarantees No guarantees => resource usage = average rate Strct guarantees => resource usage close to peak Contracts or Servce Level Agreements (SLAs) are multdmensonal quanttes Intro - 25 Resource control n wreless networks s a challenge! Wreless channel s a shared medum: smlar to legacy Ethernet CSMA/CD, but wth key dfferences Wreless transmt to the wreless channel drectly, no central pont through whch all traffc goes through Can or cannot have central control Attenuaton n the wreless medum Locaton-based contenton Smple carrer sensng not enough Can t perform collson detecton Power s an mportant resource, nfluences nterference Only recently there are mechansms for supportng QoS (80211e) Intro - 26
CDMA (Code Dvson Multple Access) Wdeband CDMA (WCDMA) most wdely adopted 3G ar nterface Based on Drect Sequence CDMA (DS-CDMA) Sgnals from dfferent mobles separated based on unque codes Transmsson rate can change between frames frequency receved power dfferent codes rate can be dfferent n dfferent frames 5 MHz 10 msec tme Intro - 27 Resource usage n CDMA: Uplnk p r g p BS RNC RNC: Rado Network Controller target bt energy to nose densty rato E b /N 0 (determnes bt error rate) spreadng bandwdth W γ = r j receved power + η rate nterference + nose assumng perfect power control g g j p p j uplnk s nterference-lmted 1 resource constrant < 1 W n uplnk + 1 γ 1 resource usage W n uplnk + 1 γ r r approxmatons for large # of moble users r γ < r γ W Intro - 28
WCDMA (3G) BS RNC Smultaneous transmssons from mobles possble Dfferentated based on code CDMA: multple access protocol Intro - 29 WLAN (80211) collson AP Smultaneous transmssons not possble Wth DSSS (Drect Sequence Spread Spectrum): same code used for all transmssons MAC (CSMA/CA) controls access/contentons to shared medum Intro - 30
WLAN (80211): Ad Hoc collson Same wth ad hoc confguraton: Smultaneous transmssons not possble But Intro - 31 WLAN (80211): Spatal reuse Lmted coverage due to path loss locaton dependent contenton Smultaneous transmsson for dstance groups (spatal reuse) Capacty of ad hoc network depends on topology Intro - 32
Multple WLANs nterference Smultaneous transmssons n dfferent WLANs possble, even n close proxmty Interference between WLANs 80211a: supports dynamc channel selecton and transmt power control (TPC) Intro - 33 Technologes WCDMA (3G) Extenson/refnement for loss senstve Extenson for power of control channels Investgaton of transent operaton va smulaton Intra-WLAN Technology: CSMA/CA (no CD), ACK, RTS/CTS Servce dfferentaton: thruput & delay, control parameters Markng: measured load/congeston, sample path Path loss results n lmted coverage Capacty depends on topology Inter-WLAN Power control to adjust qualty,coverage, and nterference between WLANs Barganng solutons? Intro - 34
Intro - 35 Possble soluton approaches Use average load over large ntervals (~mnutes) can lead to volaton of QoS Use worst case scenaros (peak rate) can lead to under-utlsaton of resources Use traffc models real world traffc not well approxmated by traffc models Use actual traffc measurements and advanced statstcal analyss procedures Intro - 36