ITTC High-Performance Networking The University of Kansas EECS 881 Control and Signalling

Transcription

1 High-Performance Networking The University of Kansas EEC 881 Control and ignalling James P.G. terbenz Department of Electrical Engineering & Computer cience Information Technology & Telecommunications esearch Center The University of Kansas rev James P.G. terbenz

2 Network Control and ignalling Outline C.1 ignalling paradigms C.2 Traffic management C.3 Path routing dynamics C.4 Monitoring and management (brief) HN-C-2

3 Network Control and ignalling Outline application application session session transport transport network network network network link link link link end system node network node end system link node network C.1. ignalling paradigms C.2. Traffic management C.3. Path routing dynamics C.4. Monitoring & management HN-C-3

4 Network Control and ignalling C.1 ignalling Paradigms C.1 ignalling paradigms C.1.0 Delay constituents C.1.1 Message and circuit switching C.1.2 Packet switching C.1.3 Fast packet switching C.1.4 Intermediate control mechanisms C.1.5 Fast circuit and burst switching C.2 Traffic management C.3 Path routing dynamics C.4 Monitoring and management HN-C-4

5 ignalling Paradigms Delay Constituents Propagation delay t p = x ij / kc distance x ij velocity of light c medium dependent constant k 0.7 Transmission delay t b = b [b] / r [b/s] witching delay t s or t r = t f + t q forwarding delay t f queuing delay t q HN-C-5

6 ignalling Paradigms Delay Constituents distance x 12 x 23 t p time t b t f +t q t r kc d IG-MG 3 Packets are parallelograms (t b thick) Messages are directed line segments (ignore t b ) HN-C-6

7 Message switching Circuit switching ignalling Paradigms 1st Generation HN-C-7

8 Messages ignalling Paradigms Message witching variable length messages transmitted on link as needed Examples? HN-C-8

9 Messages ignalling Paradigms Message witching variable length messages transmitted on link as needed Examples early data networks HN-C-9

10 ignalling Paradigms Message witching First generation (<1970) data switching technique characteristics? HN-C-10

11 ignalling Paradigms Message witching Characteristics + no setup latency HN-C-11

12 ignalling Paradigms Message witching Characteristics + no setup latency significant delay store-and-foreword delay HN-C-12

13 ignalling Paradigms Message witching Characteristics + no setup latency significant delay store-and-foreword delay HN-C-13

14 ignalling Paradigms Message witching Characteristics + no setup latency significant delay store-and-foreword delay + some multiplexing efficiency HN-C-14

15 ignalling Paradigms Message witching Characteristics + no setup latency significant delay store-and-foreword delay + some multiplexing efficiency HN-C-15

16 ignalling Paradigms Message witching Characteristics + no setup latency + some multiplexing efficiency significant delay store-and-foreword delay small messages delayed (buffered ) behind large ones HN-C-16

17 ignalling Paradigms Message witching Characteristics + no setup latency + some multiplexing efficiency significant delay store-and-foreword delay small messages delayed (buffered ) behind large ones HN-C-17

18 ignalling Paradigms Message witching Characteristics + no setup latency + some multiplexing efficiency significant delay store-and-foreword delay small messages delayed (buffered ) behind large ones HN-C-18

19 ignalling Paradigms Message witching Characteristics + no setup latency + some multiplexing efficiency significant delay store-and-foreword delay small messages delayed (buffered ) behind large ones HN-C-19

20 ignalling Paradigms Message witching Characteristics + no setup latency + some multiplexing efficiency d t significant delay store-and-foreword delay small messages delayed d r d r d t HN-C-20

21 Circuits ignalling Paradigms Circuit witching physical path established circuit state to establish and maintains path Examples? HN-C-21

22 Circuits ignalling Paradigms Circuit witching physical path established circuit state to establish and maintains path Examples early PTN early X.21 circuit switched networks modern optical WDM lightpaths HN-C-22

23 ignalling Paradigms Circuit witching Circuit signalling PTN switching technique characteristics? HN-C-23

24 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency ETUP HN-C-24

25 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency ETUP HN-C-25

26 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency ETUP HN-C-26

27 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency HN-C-27

28 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency CONNECT HN-C-28

29 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency CONNECT HN-C-29

30 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency: TT before data transfer circuit seized implication? CONNECT HN-C-30

31 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency: TT before data transfer circuit seized no multiplexing efficiency HN-C-31

32 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency: TT before data transfer circuit seized no multiplexing efficiency HN-C-32

33 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency: TT before data transfer circuit seized no multiplexing efficiency silence unusable by others HN-C-33

34 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency: TT before data transfer no multiplexing efficiency negligible switch latency HN-C-34

35 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency: TT before data transfer no multiplexing efficiency + negligible switch latency resources must be released ELEAE HN-C-35

36 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency: TT before data transfer no multiplexing efficiency + negligible switch latency resources must be released 3 4 ELEAE 13 HN-C-36

37 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency: TT before data transfer no multiplexing efficiency + negligible switch latency resources must be released 3 4 ELEAE 14 HN-C-37

38 ignalling Paradigms Circuit witching Circuit signalling characteristics setup latency: TT before data transfer no multiplexing efficiency + negligible switch latency resources must be released HN-C-38

39 ignalling Paradigms 1st Generation Comparison: Circuit vs. Message ETUP CONNECT d t d r ELEAE HN-C-39

40 How to do better than ignalling Paradigms 2nd Generation message switching? circuit switching? HN-C-40

41 ignalling Paradigms 2nd Generation How to do better than message switching: (connectionless) packet switching circuit switching? HN-C-41

42 ignalling Paradigms Datagram Packet witching Packetise message into datagrams improvement over message switching reduces queueing delay behind long messages HN-C-42

43 Connectionless ignalling Paradigms Datagram Packet witching no per flow state required to forward information but there still is state what? HN-C-43

44 Connectionless ignalling Paradigms Datagram Packet witching no per flow state required to forward information but there still is state forwarding tables other state may be used to improve performance per-flow queueing soft state flow identification to improve performance Examples? HN-C-44

45 Connectionless ignalling Paradigms Datagram Packet witching no per flow state required to forward information but there still is state forwarding tables other state may be used to improve performance Examples per-flow queueing soft state flow identification to improve performance IP (Internet protocol) CLNP (connectionless layer network protocol) [IO/IEC 8473 / ITU X.223] HN-C-45

46 ignalling Paradigms Datagram Packet witching Each datagram contains a destination address HN-C-46

47 ignalling Paradigms Datagram Packet witching Each datagram contains a destination address Each switch hop does a lookup in a forwarding table outgoing port = lookup (destination address) lecture HN-C-47

48 ignalling Paradigms Datagram Packet witching Each datagram contains a destination address Each switch hop does a lookup in a forwarding table outgoing port = lookup (destination address) Table lookup efficiency depends on: address structure (e.g. class-based IP vs. CID) address length tables length (# destinations per switch/router) Example? HN-C-48

49 ignalling Paradigms Datagram Packet witching Each datagram contains a destination address Each switch hop does a lookup in a forwarding table outgoing port = lookup (destination address) Table lookup efficiency depends on: address structure (e.g. class-based IP vs. CID) address length tables length (# destinations per switch/router) Example: IP note: IP lookup could not be done at line rate in 1980s more later HN-C-49

50 ignalling Paradigms Datagram Packet witching Connectionless signalling characteristics? HN-C-50

51 ignalling Paradigms Datagram Packet witching Characteristics + no setup latency destination address HN-C-51

52 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay HN-C-52

53 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay HN-C-53

54 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword loookup delay + multiplexing efficiency HN-C-54

55 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay + multiplexing efficiency large messages broken into packets HN-C-55

56 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay + multiplexing efficiency large messages broken into packets HN-C-56

57 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay + multiplexing efficiency large messages broken into packets + other flows interleave HN-C-57

58 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay + multiplexing efficiency large messages broken into packets + other flows interleave HN-C-58

59 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay + multiplexing efficiency large messages broken into packets + other flows interleave HN-C-59

60 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay + multiplexing efficiency large messages broken into packets + other flows interleave HN-C-60

61 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay + multiplexing efficiency large messages broken into packets + other flows interleave HN-C-61

62 ignalling Paradigms Datagram Packet witching Connectionless signalling + no setup latency store-and-foreword lookup delay + multiplexing efficiency large messages broken into packets + other flows interleave d r > d t d t d r HN-C-62

63 ignalling Paradigms Improvement of Packet over Message witching d t d t d r d r HN-C-63

64 ignalling Paradigms 3rd Generation How to do better than message switching: (connectionless) packet switching circuit switching? HN-C-64

65 ignalling Paradigms 3rd Generation How to do better than message switching: (connectionless) packet switching circuit switching: virtual circuit fast packet switching HN-C-65

66 ignalling Paradigms Connection-Oriented Fast Packet witching Connection-oriented improvement over circuit switching no physical circuit reserved connection state required performance optimisations possible to reduce setup latency fast reservations optimistic connection establishment per hop acknowledgements parameter negotiation Examples? HN-C-66

67 ignalling Paradigms Connection-Oriented Fast Packet witching Examples CON (connection-oriented network service) PPDNs (packet-switched public data networks) [IO/IEC / ITU X.25] ATM MPL modern PTN wired wireless (although migrating to IP) HN-C-67

68 ignalling Paradigms Connection-Oriented Fast Packet witching Combine benefits of datagram and circuits statistical multiplexing gains of datagrams forwarding performance circuits eliminate store-and-forward high-performance switch design details in Lecture provision of QO admission control resource reservation per connection HN-C-68

69 ignalling Paradigms Connection-Oriented Fast Packet witching Characteristics? ETUP t sig POCEEDING ETUP POCEEDING ETUP t setup POCEEDING CONNECT CONNECT CONNECT ACK t b t g ACK ACK D n t p d 1 t s t xfer d n HN-C-69

70 ignalling Paradigms Connection-Oriented Fast Packet witching Connections: establish state once to reduce per packet processing t setup ETUP t sig POCEEDING ETUP POCEEDING ETUP POCEEDING TT delay before data transfer CONNECT o amortised for long flows CONNECT CONNECT ACK expensive for transactions + high throughput possible t b t g ACK ACK D n + optimisations to reduce setup latency (later) t xfer t p t s d 1 d n HN-C-70

71 ignalling Paradigms Comparison of Characteristics Connectionless vs. connection-oriented networks major debate in 1980s and 1990s Comparison of characteristics Characteristic Connectionless Connection-oriented etup latency?? Forwarding latency Forwarding information witch state esilience to failure QO HN-C-71

72 ignalling Paradigms Comparison of Characteristics Connectionless vs. connection-oriented networks major debate in 1980s and 1990s Comparison of characteristics Characteristic Connectionless Connection-oriented etup latency none round trip Forwarding latency?? Forwarding information witch state esilience to failure QO HN-C-72

73 ignalling Paradigms Comparison of Characteristics Connectionless vs. connection-oriented networks major debate in 1980s and 1990s Comparison of characteristics Characteristic Connectionless Connection-oriented etup latency none round trip Forwarding latency address lookup VC index Forwarding information?? witch state esilience to failure QO HN-C-73

74 ignalling Paradigms Comparison of Characteristics Connectionless vs. connection-oriented networks major debate in 1980s and 1990s Comparison of characteristics Characteristic Connectionless Connection-oriented etup latency none round trip Forwarding latency address lookup VC index Forwarding information address per packet VC id per packet witch state?? esilience to failure QO HN-C-74

75 ignalling Paradigms Comparison of Characteristics Connectionless vs. connection-oriented networks major debate in 1980s and 1990s Comparison of characteristics Characteristic Connectionless Connection-oriented etup latency none round trip Forwarding latency address lookup VC index Forwarding information address per packet VC id per packet witch state forwarding tables connection id tables esilience to failure?? QO HN-C-75

76 ignalling Paradigms Comparison of Characteristics Connectionless vs. connection-oriented networks major debate in 1980s and 1990s Comparison of characteristics Characteristic Connectionless Connection-oriented etup latency none round trip Forwarding latency address lookup VC index Forwarding information address per packet VC id per packet witch state forwarding tables connection id tables esilience to failure datagrams lost* connections terminated QO?? *in theory, but higher-level (TCP) connections may time out HN-C-76

77 ignalling Paradigms Comparison of Characteristics Connectionless vs. connection-oriented networks major debate in 1980s and 1990s Comparison of characteristics Characteristic Connectionless Connection-oriented etup latency none round trip Forwarding latency address lookup VC index Forwarding information address per packet VC id per packet witch state forwarding tables connection id tables esilience to failure datagrams lost* connections terminated QO difficult connection reservation HN-C-77

78 ignalling Paradigms Connections vs. Connectionless Connections enabled fast switching in the 1980s standardised as ATM, but with serious flaws Difficult tradeoffs between two paradigms trade setup latency vs. throughput efficiency intermediate schemes? Connectionless vs. Connection Tradeoff The latency of connection setup must be traded against the reduction in the end-to-end data transfer delay due to the elimination of store-and-forward delay, and faster transmission of due to nodes capable of increased bandwidth. N-5A HN-C-78

79 ignalling Performance Efficiency and obustness ignalling complexity tradeoff between needed functionality simple state machines and message formats ignalling protocols must be robust to lost messages state machine must account for this messages should fit in single packet reliable hop-by-hop protocols add significant complexity (e.g. ATM -AAL) Efficiency of ignalling N-8B ignalling messages should be simple in coding format and fit in a single packet to minimise the latency in processing. The signalling protocol should be robust to lost messages. HN-C-79

80 ignalling Performance ound Trips Minimisation Techniques to minimise control round trips problem? HN-C-80

81 ignalling Performance ound Trips E2E ACKs E2E handshake requires TT timeout for all connect failures Alternative? t E2E ETUP ETUP ETUP CONNECT CONNECT ACK ACK HN-C-81

82 ignalling Performance ound Trips HBH upplemental ACKs E2E handshake ETUP thbh ETUP requires TT timeout te2e ETUP for all connect failures POCEEDING ETUP HBH ACKs ETUP CONNECT POCEEDING CONNECT quicker local recovery ETUP CONNECT ACK ACK CONNECT ACK ACK HN-C-82

83 ignalling Performance ound Trips Minimisation Techniques to minimise control round trips hop-by-hop acknowledgements what about failure in ETUP parameters? HN-C-83

84 ignalling Performance ound Trips Negotiation Failures pecific parameters exact match or another TT attempt Alternative? ETUP (10) EJECT (8) ETUP (8) POCEEDING ETUP (8) EJECT (5) EJECT (5) ETUP (5) ETUP (5) POCEEDING POCEEDING CONNECT CONNECT ACK ACK HN-C-84

85 ignalling Performance ound Trips ange Parameters pecific parameters ETUP (10) EJECT (8) ETUP (4-10) POCEEDING ETUP (4-8) exact match or ETUP (8) POCEEDING another TT attempt Incremental negotiation POCEEDING ETUP (8) EJECT (5) CONNECT (5) ACK CONNECT (5) ACK initiator gives range or set EJECT (5) destination ACKs capabilities ETUP (5) ETUP (5) POCEEDING POCEEDING CONNECT CONNECT ACK ACK HN-C-85

86 ignalling Performance ound Trips Minimisation Techniques to minimise control round trips hop-by-hop acknowledgements parameter ranges desired minimum acceptable desired, alternate is one TT before data flows necessary? HN-C-86

87 ignalling Performance ound Trips Data Waits for Control eparate phases control to establish path when successful: transfer data Alternative? ETUP CONNECT ETUP CONNECT ACK ACK HN-C-87

88 ignalling Performance ound Trips Overlap Control/Data eparate phases control to establish path when successful: transfer data Overlap control/data control to establish path optimistically transfer data when successful continue transferring ETUP CONNECT ACK ETUP CONNECT ACK ETUP CONNECT ACK ETUP CONNECT ACK HN-C-88

89 ignalling Performance ound Trips Minimisation Techniques to minimise control round trips hop-by-hop acknowledgements parameter ranges desired minimum acceptable desired, alternate overlap of control and data Minimise ound Trips N-6A tructure control messages and the information they convey to minimise the number of round trips required to accomplish data transfer. HN-C-89

90 ignalling Paradigms Intermediate Mechanisms Tradeoffs between packet and circuit switching? HN-C-90

91 increasing state ITTC ignalling Paradigms Intermediate Mechanisms pectrum of signalling paradigms per message forwarding (message switching) per packet datagram forwarding data-driven soft state accumulation control-driven soft state accumulation optimistic connection establishment fast reservation explicit virtual connection setup fast packet switching scheduled connection sharing burst switching explicit physical connection setup circuit switching Intermediate mechanisms benefit from extremes HN-C-91

92 ignalling Paradigms Intermediate Mechanisms Intermediate mechanisms benefit from extremes benefits of state accumulation (from connections) benefits of immediate transmission of data (from datagrams) Network Path Establishment N-II.1 The routing algorithms and signalling mechanisms must be capable of forwarding datagrams or establishing connections on sufficiently high-performance paths and with low latency to meet application needs. HN-C-92

93 Intermediate ignalling Paradigms oft tate Accumulation Features allow immediate packet send once state established: fast lookup and forwarding f o r w a r d e d t r d r Variants data driven t s d s switch detects flow from headers control driven explicit inter-switch signalling s w i t c h e d e.g. MPL HN-C-93

94 Intermediate ignalling Paradigms Optimistic Connection Establishment ignificant delay results from round trips overlapping connection setup with data transfer to reduce Overlap ignalling Messages with Data Transfer To reduce unneeded end-to-end latency, signalling messages should be overlapped with data transfer. N-6A HN-C-94

95 Intermediate ignalling Paradigms Features Optimistic Connection Establishment data sent with signalling until COMMIT returned: best effort or per hop fast reservation g t sig p r o v i s i o n a l ETUP t sig t sig COMMIT t sig r e s e r v e d HN-C-95

96 Intermediate ignalling Paradigms Burst witching Features packets group into bursts bursts scheduled in circuit t delay ETUP t sig ole t B optical networks: no photonic header processing t set Variants release: signalled or timed ELEAE in- or out-of band signalling HN-C-96

97 Multicast ignalling oot vs. Leaf Join oot join ADD Leaf join + simple 1 1 complex doesn t scale + scales JOIN oot vs. Leaf Multicast Control oot multicast control is appropriate when a single entity has knowledge of the multicast group. Leaf-controlled multicast is needed for large multicast groups. N-6B HN-C-97

98 ession control (layer 5) ession Control ignalling Efficiency set of transport layer associations (layer 4) point-to-point or multipoint (may be mixed) connection oriented or connectionless (may be mixed) HN-C-98

99 ession control (layer 5) cross-layer controls allow overlap of signalling reduction in round trips ession Control ignalling Efficiency ession Connection Interlayer Awareness ession control awareness of network layer control parameters allows the overlap of session and connection control signalling to reduce overall latency. N-4F HN-C-99

100 Latency reduction minimise round trips ession Control ignalling Efficiency user initiation parallelise session and network signalling properly locate resources user session establishment connection establishment end system network E-EQUET E-ETABLIH ETUP end system user user negotiation CONNECT data transfer HN-C-100

101 Network Control and ignalling C.2 Traffic Management C.1 ignalling paradigms C.2 Traffic management C.2.1 esource reservation C.2.2 Network-based congestion control C.3 Path routing dynamics C.4 Monitoring and management HN-C-101

102 Conflicting goals: Traffic Management Goals path protection: sufficient resources to deliver required QO to users minimise network resource use to keep costs low Network Path Protection QO mechanisms must be capable of guaranteeing bandwidth and latency bounds, when needed. N-II.2 HN-C-102

103 Traffic Management Optimality upport for mixed traffic distinct networks (1 st and 2 nd generations through 1980s) virtual network partitioning differentiated services: coarse grained service grades integrated services: fine grained traffic classes e.g. ATM-TM and intserv with VP HN-C-103

104 Traffic Management Optimality Optimal bandwidth reservation significant cost in reservation processing significant cost in state management Balance simpler admission control with some overengineering Optimal esource Utilisation vs. Overengineering Tradeoff N-5A Balance the tradeoff of optimal resource utilisation and its associated complexity and cost against the cost of suboptimal resource utilisation resulting from overengineering the network. HN-C-104

105 Network Congestion Control Motivation Limit transmission to prevent overwhelming network reservations are generally statistical congestion can occur if there is any overbooking of resources best-effort traffic must be limited to avoid congestion not all end users will behave as they should Benefits from network assistance Congestion Control in the Network Improves Performance Even though the end-to-end protocols must perform congestion control, there is substantial performance benefit in assistance from the network. N-3A HN-C-105

106 Goals Network Congestion Control Performance Goal congestion control in the network reduces control loop delay FECN CONGETION C=1 BECN CONGETION HN-C-106

107 Network Congestion Control Congestion Avoidance Congestion Avoidance reacts to impending congestion before damage is done e.g. ED (random early detection) keep queues as short as possible jitter and bursty arrivals Avoid Congestion and Keep Queues hort N-II.4 Avoid congestion by network engineering, traffic management with resource reservation, and by dropping packets. Buffers should be kept as empty as possible, with queueing only for transient situations, to allow cut-through, and avoid the latency of FIFO queueing. HN-C-107

108 Fairness Network Congestion Control Fairness vs. Complexity desirable to allow fair sharing of network difficult to discriminate well-behaved and misbehaving flows fair mechanisms substantially more complex to implement but can be done with current technology Congestion Control Fairness vs. Complexity N-2Bc The lack of fairness in simple congestion control and avoidance mechanisms must be traded against the complexity of fair implementations. HN-C-108

109 Network Control and ignalling C.3 Path outing Dynamics C.1 ignalling paradigms C.2 Traffic management C.3 Path routing dynamics Multipoint groups Node mobility C.4 Monitoring and management HN-C-109

110 Path outing Dynamics Dynamic Network Behaviour Dynamic network behaviour leads to low performance paths over time multipoint groups prune and reroute to maintain optimal topology node mobility reconfigure topology to maintain performance Dynamic Path erouting Dynamic behaviour can require adjustments to topology to maintain a high-performance path. The overhead and frequency of topology maintenance must be traded against the lack of optimality. N-2B HN-C-110

111 Path outing Dynamics Node Mobility Node mobility changes path characteristics Example d 1 +d 2 < D a intermediate node moves away latency bound exceeded: d 1 +d 2 > D a reroute path d 3 < D a d 1 d 2 d 1 d 2 d 3 HN-C-111

112 Path outing Dynamics Multipoint Groups 1 Multipoint spanning tree optimised to receiver group at a particular point in time HN-C-112

113 Dynamic behavior Path outing Dynamics Multipoint Groups 2 changes topology latency bounds may be exceeded bandwidth wasted eceivers leave group inefficient path to remaining members links carrying traffic to no receiver HN-C-113

114 Path outing Dynamics Multipoint Groups 3 Prune and reroute tree optimise to remaining group reroute to group members prune unneeded leaf hops HN-C-114

115 Network Control and ignalling Monitoring and Management C.1 ignalling paradigms C.2 Traffic management C.3 Path routing dynamics C.4 Monitoring and management HN-C-115

116 Monitoring and Management Issues Coarse granularity management not high-speed, per se Issues management must keep up with rapidly changing conditions massive amounts of data must be filtered and reduced monitoring should not interfere with high-speed flows Network Monitoring Locality N-1B Network monitoring functions must be built into the critical path to provide non-intrusive local filtering and aggregation of statistics. HN-C-116

117 Control and ignalling Acknowledgements ome material in these foils comes from the textbook supplementary materials: terbenz & Touch, High-peed Networking: A ystematic Approach to High-Bandwidth Low-Latency Communication HN-C-117