Integrating Network. Management for Cloud

Transcription

1 Integrating Network Management for Cloud Computing Services Final Public Oral Exam Peng Sun

2 What is Cloud Computing Deliver applications as services over Internet Run applications in datacenters Lower capital and operational expenses 1

3 Cloud Services are Growing Public cloud for consumer service Amazon Web Services Microsoft Azure 2

4 Cloud Services are Growing Public cloud for consumer service Amazon Web Services Microsoft Azure Enterprises out-source IT service Storage: Box, Analytics: Salesforce, Productivity: Office365, 2

5 Quality of Cloud Service Depends on Network Quality Datacenter ISP App Routing OS Hardware ISP Enterprise Network Servers Network Devices ISP 3

6 Improve Network Quality The old way cannot keep up with the growth of cloud services Deploying more devices with higher bandwidth 4

7 Improve Network Quality The old way cannot keep up with the growth of cloud services Deploying more devices with higher bandwidth The key is proper management of network resources 4

8 Examples of Network Management Solutions 5

9 Examples of Network Management Solutions Traffic Engineering 5

10 Examples of Network Management Solutions Traffic Engineering Load Balancing 5

11 Examples of Network Management Solutions Traffic Engineering Load BalancingLink Corruption Mitigation 5

12 Examples of Network Management Solutions Traffic Engineering Load BalancingLink Corruption Network Mitigation Device Configuration 5

13 Examples of Network Management Solutions Traffic Engineering Load BalancingLink Corruption Network Mitigation Device Configuration Device Power Control 5

14 Examples of Network Management Solutions Traffic Engineering Load BalancingLink Corruption Network Mitigation Device Configuration Device Power Control 5

15 Problems of Current Network Management Disjoint management of network components Low-level interfaces with network devices 6

16 Disjoint Management Systems Division of labor in pre-cloud era App OS Servers Routing Hardware Network Devices Users ISP 7

17 Disjoint Management Systems Division of labor in pre-cloud era App OS Routing Hardware Users Servers Server Provisioning Network Devices ISP 7

18 Disjoint Management Systems Division of labor in pre-cloud era Traffic Engineering App OS Routing Hardware Users Servers Server Provisioning Network Devices ISP 7

19 Disjoint Management Systems Division of labor in pre-cloud era Traffic Engineering App OS Servers Server Provisioning Routing Hardware Infrastructure Network Devices ISP Users 7

20 Datacenter Breaks Balance Coordination of server & network: More applications are built as multitier distributed systems Intra-datacenter traffic is new majority 8

21 Datacenter Breaks Balance Coordination of server & network: More applications are built as multitier distributed systems Intra-datacenter traffic is new majority Infrastructure evolution speeds up: Network architecture changes (e.g., FatTree) 8

22 Disjoint Mgmt. is Bottleneck Because: Cloud service providers have stakes in all three management areas: Server, infrastructure, traffic 9

23 Disjoint Mgmt. is Bottleneck Because: Cloud service providers have stakes in all three management areas: Server, infrastructure, traffic Great opportunity exists for consolidation Google and Microsoft on SoftWAN 9

24 Yet Another Problem: Low-level Device Interaction Hardware vendors differentiate with specialized devices Network operation: intensively uses vendor-specific APIs heavily depends on experiences 10

25 Cloud Service is Different Much more devices in datacenters than traditional networks Automation is a must 11

26 Cloud Service is Different Much more devices in datacenters than traditional networks Automation is a must Commodity devices instead of specialized hardware Homogeneity is preferred 11

27 Cloud Service is Different Much more devices in datacenters than traditional networks Automation is a must Commodity devices instead of specialized hardware Homogeneity is preferred Lower vendor dependence pays off (MS & Amazon on SoftLB) 11

28 Promising Yet Limited SDN Great way of automating traffic management with high-level programming paradigms Yet literature focus on just traffic 12

29 Summary of Problems This dissertation solves: Disjoint management of server, infrastructure, and traffic Low-level device interaction for broader scope of network management 13

30 Practical Approach 14

31 Practical Approach Identify opportunity of integration 14

32 Practical Approach Identify opportunity of integration Simple and expressive programming abstraction 14

33 Practical Approach Identify opportunity of integration Simple and expressive programming abstraction Efficient and scalable system for execution 14

34 Practical Approach Identify opportunity of integration Simple and expressive programming abstraction Deployment in real world Efficient and scalable system for execution 14

35 Practical Approach Identify opportunity of integration Simple and expressive programming abstraction Deployment in real world Efficient and scalable system for execution 14

36 Contributions Project Publication Deployment Statesman SIGCOMM 14 Microsoft Azure HONE JNSM, Vol. 23, 2015 Overture/ Verizon Sprite SOSR 15 In process with OIT 15

37 Statesman: Safe Datacenter Traffic/Infrastructure Management Other ISPs Datacenter ISP ISP App Routing Enterprises OS Hardware Servers Network Devices 16

38 Hone: End-host/Network Cooperative Traffic Management Other ISPs Datacenter ISP ISP App Routing Enterprises OS Hardware Servers Network Devices 17

39 Sprite: Direct Control of Entrant ISP for Enterprise Traffic Other ISPs Datacenter ISP ISP App Routing Enterprises OS Hardware Servers Network Devices 18

40 What Follows Brief explanation of each project Open issues Related work Q&A 19

41 Statesman: Integrating Network Infrastructure Management 20

42 Problem for Cloud Providers Multiple mgmt. solutions coexist for traffic and infrastructure mgmt.

43 Problem for Cloud Providers Multiple mgmt. solutions coexist for traffic and infrastructure mgmt. Complexity is the main problem

44 Problem for Cloud Providers Multiple mgmt. solutions coexist for traffic and infrastructure mgmt. Complexity is the main problem Development Scale & heterogeneity of devices

45 Problem for Cloud Providers Multiple mgmt. solutions coexist for traffic and infrastructure mgmt. Complexity is the main problem Development Scale & heterogeneity of devices Coordination Conflicts and safety violations

46 Conflict Core1 2 Agg A Agg B ToRs 22

47 Conflict Core1 2 Agg A Agg B Link-corruptionmitigation adjusts traffic away from Core1 ToRs 22

48 Conflict Core1 2 Agg A Agg B Link-corruptionmitigation adjusts traffic away from Core1 ToRs TE tunes traffic among links to Core1, 2 22

49 Safety Violation Core1 2 Agg A Agg B ToRs 23

50 Safety Violation Core1 2 Agg A Agg B Link-corruptionmitigation shuts down faulty Agg A ToRs 23

51 Safety Violation Core1 2 Agg A Agg B Link-corruptionmitigation shuts down faulty Agg A ToRs Firmware-upgrade schedules Agg B to upgrade 23

52 The Statesman System Network operating system Common layer to consolidate traffic and infrastructure management Resolve conflicts & safety violations 24

53 The Statesman System Network operating system Common layer to consolidate traffic and infrastructure management Resolve conflicts & safety violations Core techniques: Network state & three-view workflow State dependency model Scalable & robust system 24

54 Three-view Workflow with Network State Observed State Proposed State Target State 25

55 Three-view Workflow with Network State Application Application Solutions Observed State Proposed State Target State 25

56 Three-view Workflow with Network State Application Application Solutions Observed State What we see from the network Proposed State Target State 25

57 Three-view Workflow with Network State Application Application Solutions Observed State What we see from the network Proposed State What we want the network to be Target State 25

58 Three-view Workflow with Network State Application Application Solutions Statesman Observed State Proposed State Target State What we see from the network What we want the network to be What can be actually done on the network 25

59 State Dependency for Conflict Detection Link Device 26

60 State Dependency for Conflict Detection Link PowerState Device 26

61 State Dependency for Conflict Detection FirmwareVersion Link PowerState Device 26

62 State Dependency for Conflict Detection ConfigurationState FirmwareVersion Link PowerState Device 26

63 State Dependency for Conflict Detection ConfigurationState ConfigurationState AdminState FirmwareVersion Link PowerState Device 26

64 State Dependency for Conflict Detection RoutingState ConfigurationState ConfigurationState FirmwareVersion AdminState Link PowerState Device 26

65 State Dependency for Conflict Detection PathState RoutingState ConfigurationState ConfigurationState FirmwareVersion AdminState Link PowerState Device 26

66 Statesman System Observed State Proposed State Target State 27

67 Statesman System Storage Service Observed State Proposed State Target State 27

68 Statesman System Storage Service Observed State Proposed State Target State Monitor 27

69 Statesman System Storage Service Checker Observed State Proposed State Target State Monitor 27

70 Statesman System Storage Service Checker Observed State Proposed State Target State Monitor Updater 27

71 Deployment Overview Operational in Microsoft Azure since October 2013 Cover 10 DCs of 20K devices 3 production management solutions built and running 28

72 Case: Resolve Conflict Inter-DC TE & Firmware-upgrade DC = Data Center BR = Border Router BR 3 DC 2 BR 4 BR 1 DC 1 BR 2 BR 8 DC 4 BR 7 BR 6 BR 5 DC 3 29

73 30

74 30

75 Firmware-upgrade acquires lock of BR1 30

76 TE fails to acquire lock, and moves traffic away 30

77 TE fails to acquire lock, and moves traffic away 30

78 BR1 firmware upgrade starts 30

79 BR1 firmware upgrade starts BR1 firmware upgrade ends. Lock released. 30

80 BR1 firmware upgrade starts TE re-acquires lock, and moves traffic back 30

81 BR1 firmware upgrade starts TE re-acquires lock, and moves traffic back 30

82 Statesman Summary Programming abstraction Three-view network state model Efficient and scalable system Automatic and safe infrastructure management system Deployment Operational in Microsoft Azure worldwide SIGCOMM

83 Hone: Combining End-host and Network for Traffic Management 32

84 Problem Server Traffic Management Solutions App OS Network Traffic management is limited by the scope of network devices Coarse granularity & limited view 33

85 The Hone System Bring end hosts into traffic mgmt. 34

86 The Hone System Bring end hosts into traffic mgmt. Core techniques: Access of socket & transport layers 34

87 The Hone System Bring end hosts into traffic mgmt. Core techniques: Access of socket & transport layers Expressive three-stage programming framework 34

88 The Hone System Bring end hosts into traffic mgmt. Core techniques: Access of socket & transport layers Expressive three-stage programming framework Efficient system with parallel execution of partitioned program 34

89 Programming Model Measure Stage Analyze Stage Control Stage Framework for each stage Programmable body of each stage Focus on measurement and analysis 35

90 Hone System Host App Server OS Network 36

91 Hone System Host App Server OS Hone Agent Network 36

92 Hone System Controller Mgmt. Program Hone Runtime System Host App Server OS Hone Agent Network 36

93 Hone System Controller Mgmt. Hone Runtime Program System Host App Server OS Hone Agent Network 36

94 Hone System Controller Host Host Local Exe. Plan Global Hone Runtime System Net Local Exe. Plan Exe. Plan App Server OS Hone Agent Network 36

95 Hone System Controller Global Hone Runtime System Exe. Plan Host Host Local App Hone Exe. Plan Agent Server OS Net Local Exe. Plan Network 36

96 Evaluation Built multiple traffic management solutions on Hone Show distributed rate limiting here Limit total bandwidth across all instances of a tenant in a public cloud 37

97 DRL on HONE Host-side execution: Measure Calculate throughput Aggregate among connections Controller-side execution: Aggregate among hosts Generate new rate-limit policies 38

98 Distributed Rate Limiting 39

99 Distributed Rate Limiting 39

100 Distributed Rate Limiting 39

101 Distributed Rate Limiting 39

102 Distributed Rate Limiting 39

103 Hone Summary Programming abstraction Access to fine-grained data in servers Three-stage framework Efficient and scalable runtime Deployment Integrated into product of Overture for Verizon Business Cloud Springer JNSM Volume 23,

104 Sprite: Bridging Enterprise and ISP for Inbound Traffic Control 41

105 Problem for Enterprises: Inbound Traffic Engineering Service 1 ISP 1 Internet ISP 2 Enterprise Network Service 2 ISP 3 42

106 Problem for Enterprises: Inbound Traffic Engineering ISP 1 Internet ISP 2 Enterprise Network Service 2 ISP 3 42

107 Problem for Enterprises: Inbound Traffic Engineering ISP 1 Internet ISP 2 Enterprise Network Service 2 ISP 3 42

108 Mechanism of Sprite Enterprise Network ISP 1 YouTube User Other ISPs ISP 2 VoIP 43

109 Mechanism of Sprite Enterprise Network /24 ISP 1 YouTube User Other ISPs ISP 2 VoIP /24 43

110 Mechanism of Sprite Enterprise Network /24 ISP 1 YouTube User Other ISPs ISP 2 VoIP /24 43

111 Mechanism of Sprite /24 Enterprise Network Src: User ISP 1 Other ISPs YouTube ISP 2 VoIP /24 43

112 Mechanism of Sprite /24 Enterprise Network Src: User ISP 1 Other ISPs YouTube Src: ISP 2 VoIP /24 43

113 Challenges Naïve solution All done by the border router 44

114 Challenges Naïve solution All done by the border router Need a distributed solution 44

115 Challenges Naïve solution All done by the border router Need a distributed solution Control-plane scaling 44

116 Challenges Naïve solution All done by the border router Need a distributed solution Control-plane scaling Data-plane scaling 44

117 Challenges Naïve solution All done by the border router Need a distributed solution Control-plane scaling Data-plane scaling Need a simple management interface 44

118 Three Tiers of Abstractions Abstraction Example 45

119 Three Tiers of Abstractions Abstraction Example High-level Policy 45

120 Three Tiers of Abstractions Abstraction High-level Policy Example <BioDept, Analytic> BestLatency 45

121 Three Tiers of Abstractions Abstraction High-level Policy Example <BioDept, Analytic> BestLatency Networklevel Rule 45

122 Three Tiers of Abstractions Abstraction High-level Policy Example <BioDept, Analytic> BestLatency Networklevel Rule < /16, /24> ISP1 45

123 Three Tiers of Abstractions Abstraction High-level Policy Example <BioDept, Analytic> BestLatency Networklevel Rule < /16, /24> ISP1 45

124 Three Tiers of Abstractions Abstraction High-level Policy Example <BioDept, Analytic> BestLatency Networklevel Rule < /16, /24> ISP1 Per-flow Rule 45

125 Three Tiers of Abstractions Abstraction High-level Policy Example <BioDept, Analytic> BestLatency Networklevel Rule < /16, /24> ISP1 Per-flow Rule < , > SNAT to

126 Case: Dynamic Perf-driven Balancing Move users connections among ISPs for best per-user performance Live Internet experiment on AWS VPC and PEERING 10 users watch movies on YouTube 46

127 User YouTube Throughput 47

128 Sprite Summary Direct and fine-grained inbound traffic control Scalable solution Scaling control and data planes Evaluation In collaboration with OIT SOSR 15 48

129 Open Issue #1 Statesman + Hone Merge Hone into Statesman Joint server, traffic, infrastructure mgmt. Possible exploration State abstraction for server data Dependency of server and network states Safety invariants involving servers 49

130 Open Issue #2 Transactional Statesman Transactional semantics for conflict resolution Possible exploration Grouping semantics Condition semantics 50

131 Open Issue #3 Hone for Multi-tenant Cloud Loosen the assumption of having access to the hosts OS Possible exploration Infer stats in guest OS from data in hypervisor 51

132 Related work 52

133 Statesman Related Work Work What they do Statesman SDN works Onix Pyretic Corybantic Centralized control of flows Single repository of network stats Target at flow management Tight cross-application proposal evaluation Wider spectrum of management applications Three-view network state model Wider spectrum of applications Loose coupling FlowVisor Virtual topology slicing Network state model 53

134 Hone Related Work Work What they do Hone Network Exception Handler Use hosts only as software switches Go deeper into socket layer Gigascope Extend SQL Use functional language to construct Chimera Specific application (intrusion detection) Support various management solutions MapReduce Naturally parallelizable data Data inherently associated with collection point 54

135 Sprite Related Work Work What they do Sprite BGP AS-path prepending Works on Internet rearchitecture Tunnel-based works Tune BGP configurations to influence neighboring ASes Clean-slate design in routing or hosts Two ends cooperate to control traffic Direct control from inside enterprise networks Incrementally deployable Only need the client side to act 55

136 Thanks! Q&A 56