1.

What is the design principle called when the path for a flow does not depend
on the current traffic matrix?

- Oblivious routing

2. What does VL2 use to spread traffic evenly over all available paths?

- Valiant Load Balancing

3. What is assigned to the intermediate switches in the VL2 routing

implementation?

- Anycast address

4. In VL2 routing, what wrapping allows the forwarding of packets to destination

switches?

- An inner header with the destination forwarding address

5. What type of addresses are location independent and provide an illusion of a

single big Layer 2 switch?

- Application addresses

6. What maintains a directory server mapping application level addresses to

current locators in VL2?
- Indirection or Virtualization layer

7. What does ECMP allow in VL2 routing?

- Selection among shortest paths for load balancing

8. What ensures performance uniformity in VL2 using Clos networks?

- Non-blocking network

9. What is needed on hosts for L2 broadcast and multicast in VL2?

- A 'layer 2.5' shim agent running

10. What is meant by location independent addressing in VL2?

- AAs are independent of VM's physical location

11. How is the application address mapped to current locators in VL2?

- Through a directory server maintained by the Indirection or Virtualization layer

12. What is important for performance isolation among tenants in VL2 according
to the text?

- TCP backing off to match the destination's rate

13. What principle allows fast load balancing in VL2?

- Security

14. What will be studied in the lecture on Geo-distributed cloud data centers?

- Interaction of data centers with users and with other data centers.

15. Name the data center interconnection techniques discussed.

- Traditional schemes like MPLS, Google's B4, and Microsoft's Swan.

16. Why is internet connectivity crucial for application service performance?

- Connectivity for services depends on the internet.

17. Why does Google require an extensive infrastructure with multiple locations?

- For better data availability, load balancing, reduced latency, local data laws compliance, and
hybrid operation.

18. What was the focus of the study based on five Yahoo data centers from
2011?

- Significant inter-data center traffic.

19. In terms of traffic flows, what percentage is traffic between data centers
compared to clients?

- 10% to 20% - flows between data centers can carry more data.

20. Why are networks between data centers and end-hosts different?

- Persistent dedicated, high-speed connectivity between a set of end-points.

21. What is the traditional approach for traffic engineering in networks discussed?

- MPLS (Multiprotocol Label Switching) Network.

22. What is emphasized as expensive in Microsoft's perspective?

- The WAN design incurring annual costs of millions of dollars.

23. What is the purpose of using different IP subnets?

- To act as topological locators for routing across a network.

24. Why is it challenging to change IP addresses of live running services?

- Changing IP addresses can disrupt services and require reconfiguration.

25. How does cloud computing address performance uniformity?

- Ensures consistent performance and latency for virtual machines wherever they are.

26. What is micro-segmentation in network security?

- Fine-grained control over data flow between applications or tenants.

27. Why is network semantics important in a software-defined network?

- Ensures functionality beyond just Layer 3 routing, including Layer 2 services.

28. What is the concept of network virtualization in multi-tenant data centers?

- Creating virtual networks over a shared physical infrastructure to isolate tenants.

29. What are the key needs identified in the VL2 case study?

- Agility, location-independent addressing, performance uniformity, security, and network

semantics.

30. How does the VL2 design ensure high throughput?

- By implementing a nonblocking fabric with high throughput for any traffic matrices.

31. What is the design result of implementing a Clos topology?

- A non-blocking topology that enables scaling out instead of up.

32. Describe the failure characteristics analyzed in the VL2 case study.

- Examined failures resulting in 0.4% unresolved issues and 0.3% loss of redundancy.

33. What physical topology did VL2 implement?

- A Clos network with interconnected aggregation and intermediate switches.

34. What is the purpose of using Linux containers?

- To create an environment similar to a standard Linux without separate kernel.

35. Explain the performance advantage of Linux containers over VMs.

- Containers offer near bare metal performance and faster provisioning times.

36. How do containers achieve isolation similar to VMs?

- Containers isolate applications using separate namespaces.

37. What is Docker in the context of Linux containers?

- Docker is a container manager for multiple containers.

38. How does Docker ensure isolation between applications?

- By using separate namespaces, preventing resources access between applications.

39. Explain the networking setup in Docker containers.

- Each container gets a virtual interface connected through a virtual ethernet bridge.

40. What role does NAT play in Docker networking?

- NAT provides external network connectivity in Docker containers.

41. How does the hypervisor improve network performance for VMs?

- Hypervisor runs a virtual switch to facilitate network communication.

42. What is the primary factor influencing CPU usage in packet processing?

- Packet filters and forwarding logic can make packet processing CPU-intensive.

43. Explain the challenges of packet I/O processing on CPUs.

- Moving packets from NIC buffers to OS buffers requires CPU interrupts.

44. What is the limitation on time for making forwarding decisions at 10Gbps?

- At 10Gbps, there's only a 67ns interval between packets for forwarding decisions.

45. How has recent engineering improved packet I/O performance on CPUs?

- Engineering efforts have made it possible to handle packet I/O even at high speeds.

46. Where do the switch routing decisions lie in Open vSwitch?

- User space

47. What type of logic is desirable to keep in the kernel in Open vSwitch?

- Simple

48. What is optimized for processing network updates in Open vSwitch?

- User space

49. In Open vSwitch, what handles packet forwarding largely?

- Kernel

50. What is the purpose of creating a hash table based cache in Open vSwitch?

- To optimize packet forwarding

51. What does Mininet create for realistic virtual networks?

- Virtual network emulator

52. What is emulated by Mininet as an OS level process?

- Host

53. What CLI does Mininet provide for interacting with the network?
- Command Line Interface (CLI)

54. What does MiniEdit in Mininet provide as a GUI application?

- Topology generation

55. What does VND in Mininet allow for automating?

- Creation of scripts

56. What are the challenges of using specialized hardware for networking VMs?

- Sharing NIC resources and isolating virtual machines.

57. Describe the main idea behind the hardware-based approach for networking
VMs.

- Utilizing NICs designed for packet forwarding instead of CPUs.

58. How does SR-IOV address the challenges of sharing NIC resources and
isolating virtual machines?

- By providing virtual functions for each VM mapped to NIC hardware.

59. Explain how SR-IOV facilitates bypassing the hypervisor for packet
transmission.

- Through DMA, which moves packets directly to VM memory.

60. What are the benefits of using SR-IOV for networking VMs?

- Higher throughput, lower latency, and lower CPU utilization.

61. What are the downsides of using SR-IOV for networking VMs?

- Trickier live VM migration and less flexible forwarding.

62. What is Open vSwitch designed for in cloud computing?

- Flexible and fast-forwarding of network traffic.

63. How does Open vSwitch partition tasks between user space and kernel
space?

- By dividing tasks between user and kernel space for efficient development.

64. What differentiates the software-based approach from the hardware-based

approach in VM networking?

- Software provides more flexible forwarding and live VM migration.

65. What can directory servers do in a system?

- Directory servers can orchestrate application locations and control communication policy.

66. According to the course notes, what does VL2 refer to?

- VL2 is a case study in the area of network virtualization.

67. What approach to virtualization was introduced by Teemu Koponen et al.?

- Teemu Koponen et al. introduced the NVP approach to virtualization.

68. Which company acquired Nicira startup, as per the notes?

- VMware acquired the Nicira startup.

69. Describe the network virtualization platform introduced by Teemu Koponen et

al.

- It replicates arbitrary topologies using OpenFlow forwarding tables.

70. In the network hypervisor, what type of network can tenants build?

- Tenants can build any standard layer 3 network.

71. What does the network virtual service consist of?

- It is a sequence of data path elements represented by OpenFlow tables.

72. What is the challenge related to performance in network virtualization, as

discussed in the notes?

- There is a large amount of state to compute.

73. How does the solution address the challenge of pipeline processing in virtual
switches?

- The first packet of a flow goes through the full pipeline.

74. Discuss the challenges in multi-tenant data centers as mentioned in the

notes.

- Challenges include agility, location-independent addressing, performance uniformity, security,

and network semantics.

75. What were the key topics of discussion in the lecture on software-defined
networks as per the notes?

- The need for software-defined networks, key ideas, and challenges.

76. Who is the associate professor mentioned in the course notes?

- Dr. Rajiv Misra is the associate professor at IIT Patna.

77. Why do networks typically run around 30% utilization?

- To have enough headroom for inefficiencies like local greedy choices.

78. What is a challenge with achieving network-wide fairness using MPLS?

- Link-level fairness only; needs global network view.

79. How does service diversity help achieve high WAN bandwidth utilization?

- Some services tolerate delay and fill in bandwidth fluctuations.

80. What is the concept of centralized TE using SDN in WAN context?

- Gathering network state info for centralized traffic flow decisions.

81. Why is exact linear programming considered too slow for dynamic network
traffic reallocation?

- Complexity due to multitude of services and quick decision requirements.

82. How does dynamic reallocation of bandwidth pose a challenge?

- Requires continual high-priority traffic decisions with quick algorithms.

83. What are the important links for Mininet?

- mininet.org, github.com/mininet, github.com/mininet/mininet/wiki/Documentation

84. How does SR-IOV hardware-based approach sacrifice flexibility?

- It sacrifices flexibility and forwarding logic for line rate performance.

85. What compromise does Open vSwitch make in the software-based

approach?

- It compromises worst-case performance for forwarding flexibility.

86. According to the lecture, what are the two methods of virtualization
discussed?

- Docker-based and Linux container-based virtualization.

87. Which two approaches are discussed to address networking VMs?

- Using SR-IOV specialized hardware or Open vSwitch software approach.

88. Who authored the performance comparison of VMs and Linux containers?

- W. Felter, A. Ferreira, R. Rajamony, J. Rubio

89. What are the complexities that SDN addresses in traditional networks?

- Complexity of existing networks and lack of clear programming APIs.

90. How are configurations interpreted in traditional network devices?

- Configuration files are interpreted by router software implementing protocols.

91. What is the need for SDN in traditional networks?

- To innovate new networks efficiently due to complex traditional setups.

92. What does policy encapsulation into device software imply in a traditional
network?

- Network routing policies are embedded in distributed protocol implementations.

93. What are the two methods of server virtualization discussed in the lecture?

- Docker based and Linux container based.

94. What hardware-based approach is used to network VMs in server

virtualization?

- SR-IOV (single-root I/O virtualization).

95. How does server virtualization benefit cloud computing infrastructure?

- Enables rapid deployment, VM multiplexing, fast VM migration.

96. What role does a Hypervisor play in server virtualization?

- Manages physical hardware and provides emulated view to VMs.

97. What networking component is virtualized by the Hypervisor in server

virtualization?

- Physical network interface card (NIC).

98. What is the purpose of a virtual switch in the Hypervisor during server
virtualization?

- To connect VMs to each other and the external network.

99. In Docker-based virtualization, what benefit does Docker Engine offer?

- Abstraction and automation of OS-level virtualization.

100. How does Docker container differ from a Virtual Machine (VM)?

- Docker container shares kernel, more efficient than VM.

101. How does each VM operate in VM-based virtualization application?

- Each VM runs its own guest OS where application processes run.

102. What fault-tolerance mechanism is used in the software solution?

- Paxos

103. How is the scalability of the system achieved?

- Using a hierarchy of controllers

104. What problem does the software solution solve in 0.3 seconds?

- Trafficking

105. What is the top-level controller in the hierarchy of controllers?

- Global controller

106. What is the SDN gateway's role in the system?

- Interact with global controller

107. How are traffic engineering decisions made at a global scale?

- Based on flow groups

108. What kind of links connect each pair of sites in the architecture?

- Massive capacity links

109. What technique does Microsoft's Swan use to avoid congestion?

- Optimizing traffic flow changes

110. Which three data center interconnection techniques are discussed?

- (i) MPLS (ii) Google's B4 and (iii) Microsoft's Swan

111. What is the purpose of using OSPF and IS-IS in a network?

- To flood information about the network's topology to all nodes.

112. How do OSPF and IS-IS help in traffic engineering?

- They spread information about bandwidth usage on different links in the network.

113. How are tunnels provisioned in a network with OSPF and IS-IS?

- Routers set up tunnels along the shortest path with sufficient capacity.

114. What is the role of MPLS in the context of tunnel provisioning requests?

- MPLS supports priorities, allowing higher priority flows to displace lower ones if needed.

115. In MPLS, who reads the packet header along the path?

- Only the ingress and egress routers read the packet header.

116. Why is the protocol named Multi-Protocol Label Switching (MPLS)?

- Because it makes forwarding along the path very simple.

117. What is the main issue addressed in the text regarding network inefficiency?

- Inefficient usage of expensive bandwidth due to peak traffic provisioning.

118. How can networks be optimized to address inefficiency in bandwidth usage?

- By differentiating traffic by service, such as latency-sensitive and background traffic.

119. What are the key ideas of Software-defined Networking (SDN)?

- Division of Policy and Mechanisms, Low-level interface, Logically centralized controller.

120. What is the purpose of a logically centralized controller in SDN?

- To express goals in one location, simplify switching gear, and centralize intelligence.

121. What are some examples of SDN controllers mentioned in the notes?

- NOX, 4D architecture, Ethane, OpenFlow.

122. How did the evolution of SDN focus on Flexible Data Planes?

- Drive network flexibility, implement label switching or MPLS for traffic optimization.

123. What was the concept of 'Centralized Control' in the evolution of SDN?

- Centralized computing for routing, logically separated decision planes, and enforcing network
policies.

124. What is the purpose of edge rate limiting in cloud data centers?

- To enforce flow rates at the edge of the network.

125. What are some unique characteristics of Google's B4 WAN?

- Massive bandwidth requirements, elastic traffic demand, and full control over edge servers
and network.
126. How is routing control implemented within Google's B4 WAN?

- Routing control is moved to a software router using Quagga software switch with OpenFlow.

127. What is the role of the Traffic Engineering server in Google's B4?

- The TE server collects topology and bandwidth info, manages flow demands, and allocates
flows to data centers.

128. Why did Google opt for custom-built hardware for B4 in 2011?

- Due to the limited availability of software-defined networking switches on the market.

129. Where does most intelligence reside in Google's B4 design?

- Most intelligence resides in the software.