1
Q
What steps need to be done in order to implement mobility in networking?
A
Detect mobility: change of network attachment point; Adding, removing, dis- and reconnecting
Dealing with mobility:
- Rendezvous for new communication relationships (client vs server vs peer mob)
- Handover of existing comm relationships (one vs two mobile nodes; double-jump problem)
- Recovery
- > Control plane protocol for handling mobility
- > Opt: Data plane encapsulation/fwd protocol
2
Q
Name 4 examples for mobility actors
A
- Mobile node: tx/rx packets, sometimes accepts connections
- Corresponding node: peer of mobile node (might also be mobile)
- Network nodes: routers, access points, switches
- Indirection nodes: intermediaries, proxies, agents
3
Q
What issues does mobility imply into networking?
A
- Adds states and complexity to nw: Als, agents, intermediaries require secure signaling to update, nodes need to support this
- Single points of failure
- Introducing potential bottlenecks (congestion in node or access link)
- Indirections may introduce path stretch
- Security issues
- Complexity for endpoints and applications (legacy apps)?
4
Q
What is path stretch?
A
path stretch = mobile node path length / shortert path length
5
Q
Why is IP address = identifier a problem for mobility? How can we deal with this?
A
Historically, IP addresses were used to identify clients
- authentication purposes
- security contexts
- find transport control blocks
- establish (multiple) connections with each other
- > invalidations when changing IP addr
We can either avoid changing the IP address or deal with a changing IP addr
6
Q
Describe issues with clan mobility
A
- Break-before-make (simplicity)
- client decides when to handover (sig strength, missed beacons, tx retires); stability vs client agility
- to which ap?
- initiate handover
7
Q
How does a plan handover work?
A
- probe for new APs and pick one
- Dissociate from present
- Auth with new
- Associate with new
oldAP determines that A moved away. newAP tells oldAP that he is responsible for A and oldAP should send over buffered msgs; oldAP should update its routing table
8
Q
Describe how mobileIP works
A
- Every network has a Home/Foreign agent (discovery via adv/sol msgs).
- Mobile node determines if home or foreign.
- Foreign: register care-of-address with HA (via FA)
- > when returning home, deregister the latest care-of address
9
Q
How do mobile IPv4 and mobile IPv6 differentiate?
A
- No foreign agent
- HA changes prefix from home prefix to foreign prefix
- uses neighbor discovery & reachability checks
10
Q
What are return ratability checks?
A
- Used after binding updates to the correspondent node
- ensures that both paths via Hader and Care-of Addr reach the same nodes in both directions
- Send Home test init + Care-of test init at same time
- replied to with minimal overhead by CN
11
Q
What improvements does Hierarchical MobileIP offer?
A
Reduce number of RTTs -> reduce latency for binding updates to HA and CNs
by: create and maintain tunnel between CN and MN.
Regional Care-of address (mobile anchor point). Fast updates between router and anchorpoint
12
Q
What is the host identity protocol?
A
Provides identifiers across interfaces for computing platforms which handles association setup.
4-way handshake
- authentication, establishing IPSec SAs, DH keys, DoS protection
- Dynamic rekeying during exchange
- Support for multi-homing and mobility
- Initial contact via DNS
13
Q
What is the Host Identifier in HIP?
A
- Public/private key pair
- authenticates coupling/decoupling
14
Q
What is the Host Identity Tag and the Local Scope Identifier in HIP?
A
HIT: 128bit representation of HI
LSI: 32bit locally generated identifier.
Looks like it was taken from IPv4 addr space
15
Q
How does IP addr updates work in HIP?
A
1 send remote adder parameter to peer
2 wait fro SPI from peer
3 transmit data using SPI
16
Q
How does HI resolution work in HIP?
A
- Initial use of DNS: map dns name to IP addr and HI
- Send IP packet to target, renegotiate bindings
- provide remote address updates during operation
17
Q
What benefits does HIP add?
A
- Rendezvous servers may also help interworking with non HIP systems
- provide fixed point of contact
- perform packet forwarding
- HIP provides third namespace (after IP and DNS)
- IP addr independence naming of computation platforms
- incl Multi-homing, mobilitys
- identifiers across NATs and middleboxes
- security
18
Q
Name the 3 distinct functions of mobility
A
- Anchroing function
- Internetwork location management
- Forwarding management
19
Q
Name some issues that arise in wireless networks
A
- Congestion vs bit error losses: L4 Protocol cannot rely on traditional congestion indicators
- sudden changes in link performance (bitrate, latency, error rate until link loss)
- Disconnections will cause series of packet losses
20
Q
What is freeze tcp?
A
A mechanism to prevent the sender from sending packets in the first place when a timeout occurs.
- It works by setting rwin = 0 in tcp header before disconnection
- Reset after disconnection is over
- Prevent messing with congestion control
21
Q
What are session continuity extensions for SSH and TLS and how do they work?
A
Extensions that deal with disconnect in encrypted transport layer photos.
Basic design goals are:
- end2end operation
- incremental deployment
- minimize endpoint notfications
- don’t optimize for fast handover
Mechanisms:
- in-band signaling using ssh/tls framing
- secure signaling
- explicit session layer ACKs for reliability and sync
- Controlled session teardown
22
Q
What is RCP?
A
It is a receiver-driven transport protocol.
- loss recovery considering the wireless link
- loss differentiation (congestion vs error) for CC
- energy efficiency considerations
23
Q
Name some properties how HTTP can deal with mobility
A
- Client-server op: Mobility is easily supported for mobile clients
- Stateless operation
- uniform interface with self-descriptive messages
- often idempotent operations (repeated execution yields same results)
- proxies as essential part of system
24
Q
What are performance enhancing proxies (PEPs)?
A
Proxies that
- operate across layers (l3 & l4)
- operate in pars: allows the usage of internal protocols between PEPs (better performance, compatibility)
- focus on common app protocols (HTTP, DNS)
25
How can the client be informed to use a PEP?
1 proxy settings in app
2 generic intermediary settings in app (SOCKS)
3 Transparent capturing of connections (terminate
26
Name some typical use cases for PEP functions
1 Data volume reduction (caching, compression, adaptation - e.g. bundling)
2 Latency reduction (caching, prediction and prefetching)
3 Robustness (hiding and dealing with path/conn failures)
27
Name some possibilities to predict user traffic?
User behavior:
- browsing behavior by individuals or groups
- Build dependency and expectation graphs
- For embedded links and indenpendent sites
Browser behavior:
- more predictable (will fetch embedded scripts and objects)
28
What disadvantages does prefetching have?
may increase load
may miss resources
- JS parsing header
- ephemeral content
- content personalization
29
What is delegation in mobile network performance optimization?
Pushing functions towards the server
e.g. DNS lookup:
• Prefetching on the server side implies DNS lookups
• Shorter RTT, fewer losses -> faster response
• URIs in the HTML may be replaced and/or DNS results pushed
• Using configured HTTP proxies also delegates the lookup
30
Name some problems regarding robustness of PEPs
- Split connection approaches with Paris of PEPs allow shielding apps from temp disconnects
- App timeouts for completing an op
- PEPs need to entertain apps while they wait
31
What is the main security issue of PEPs?
PEPs at app layer might fail with https
They need to look
- at requests for caching
- responses for parsing
- content for transformation or compressions
32
What are MANETs and VANETs?
Mobile Ad Hoc Networks and
| Vehicular Ad Hoc Networks
33
What are some specifics about MANETs?
- Hosts connect via short range radio
- All hosts are routers, all routers are mobile -> Address tied to host not topological location
- No stable topology -> cannot solve mobility through indirection
34
Name some scenarios for MANETs
- Conferencing
- Home networking
- Emergency services
- Ubiquitous computing
- Sensor networks
- Military tactical networks
- Vehicles
35
Name some problems in the layering for MANETs
Network layer: New routing mechanisms required
| Link layer: Need medium access control w/o central AP
36
Name and explain two traditional algorithms for finding next hops
Link state algorithm: Each node collects own link state and periodically floods them into network. Each router knows much of the topology and thus can make path decisions. High store requirement
Distance Vector algorithms:
Nodes advertise their chosen routes to known destinations (announces complete routing table). Correct operation depends on the correctness of their neighbors routing tables. Prone to routing loops and dead-ends
37
Describe some differences between MANETs and traditional routing
- Every node is a router and fwds packets
- Topologies are highly dynamic
- Larger number of redundant paths
- Nodes have single radio face rather than multiple isolated links
- Link properties are highly variable and can be poor
- Nodes have limited resources
38
Explain shortly how Optimized Link State Routing works
Each node detects own links and floods them into network. Every node can build a full network graph (-> shortest path).
Routing is done hop-by-hop: each node calculates the shortest path
39
How can OLSR route advertisement be optimized?
Reduce information propagation
Only get information from routes that:
- are not direct neighbours
- are not the same node multiple hops away from multiple peers
- > Multipoint relay set
40
How does the Ad hoc on-demand distance vector algorithm (AODV) work?
Route discovery:
- Floods route request
- Nodes forwarding the RREQ to construct reverse path
- When destination receives RREQ -> sends RREP
- Node on the reverse path create an entry in their routing table
Route maintenance:
- Node along the path detects a failure sends a RERR. Source initiates new RREQ
41
What is the difference between AODV and DSR?
Dynamic Source Routing: the source wants to learn the full path
AODV: establish routing tables
42
How can routing be scaled in ad hoc networks?
Detect clusters and exploit them for scaling.
3 types:
- Cluster Heads: Maintains cluster relationships
- Cluster Gateways: Provides connectivity to other clusters
- Cluster Members
43
What can clustering be used for?
Transmission management
Backbone formation
Routing efficiency
44
What are the 3 main areas of application of VANETs?
- Safety: cooperative fwd and intersection collision warning; traffic light violation warning
- Transport efficiency: Traffic light optimal speed advisory
- Information and entertainment: Remote wireless diagnostics, content and internet services
45
What is a dynamic network?
A dynamic network consists of contacts, nodes and time. These entities are connected and the connections my vary according to time.
46
Name 3 routing mechanisms
Forwarding: pass single copy of msg to contacts
- > requires full path calculation
- > need to know contacts
Replicating: Create multiple copies of msg
-> redundancy if one might get lost
Coding: transform messages before transport
-> Erasure coding (fwd err corr), network coding
47
What is epidemic routing?
Flooding: If we're not sure which next-hop to use, use all. More fine-grained: use some and limit number of allowed forwards.
Reliability vs efficiency: More nodes increases success chance, more forwarding leads to worse efficiency
Scalability issues, expensive in terms of BW
48
What is spray and wait?
1- Spread messages into network until max num of copies is achieved
(spray pattern can be controlled (e.g. always give half of remaining copies))
2- If none of the messages reached ist, wait until a node carrying a msg meets dst
49
What are some improvements to epidemic routing (made by estimations)?
- Drop least encountered: prefer frequently seen relays
- PROPHET: prob estimate for succ delivery
- Context-aware routing: store bundles in nodes w/ highest delivery prob
- Meets and visits: builds and propagates a picture of overall connectivity
50
What is R^2CP?
It is an extension for RCP which introduces mobility extensions
- seamless handovers
- server migration
- bandwidth aggregation across multiple links
51
What is the main difference between loss in a wired network and in a wireless network? How does this mess with TCP CC?
In wired network, packet loss can be an indicator for congestion.
In wireless networks, random packet loss occurs more frequently (including single packet loss).
Since TCP CC (loss-based) assumes that the network is congested if a packet gets lost, tcp will detect congestion more often in wireless networks although only one single random packet was lost
52
What is the main difference between TCP and RCP?
In RCP, congestion & flow control is handled in the receiver then in the sender.
53
What is multi path TCP?
Multihoming support for tcp
- Allows adding/removing endpoints (IP addresses) to a TCP connection
- May spread the load across connections (resource pooling)
- can failover from one address to other
(like our protocol)
54
What is SCTP?
Stream Control Transmission Protocol
- allows multiple interface with stream multiplex
- chooses primary address per endpoint
- can failover from one address to another
- numerous mobility extensions suggested to add/remove addresses dynamically
55
How can packet buffering optimize wireless connections?
Snooping in AP
| looks at tcp flow and caches data segments and retransmits them on DUP ACKs
56
How can header compression optimize wireless connections?
It offers several connection states:
- created (and restored on the fly); timed out
- used to predict header fields
- replace headers by short context identifier
- transmit the header parts deviating from the expectation
- losses and expectation mismatches can be repaired within 1 RTT
57
How can splitting a TCP connection enhance wireless performance?
2 independent TCP connections:
mobile <> AP (+PEP) <> target
wireless: shorter RTT, less predictable, fast reaction
wired: longer RTT, more stable but not fast reaction required
PEP can be introduced
58
What is the basic idea of MOWGLI? What is a problem with this?
Replace TCP connection on wireless link with more efficient protocol (e.g. via MSOCKS) to the AP. Switching attachment point would result in routing back to the original AP.
A problem could be that when the client moves, it moves away from the optimal intermediary which might result in worse performance. I-TCP and Mobile TCP can fix this by migrating connection to another middlebox
59
What is an advantage of using two intermediaries?
- minimizes impact on both endpoints: clients and servers can run without modifying OS or apps
- allows for joint operation across different mobile nodes: coupling CC, efficient compression, higher layer performance enhance
- match for mobile networks where access routers in vehicles need to be deployed anyway
60
Name some meta issues with L4 optimizations
- Full isolation of wireless link from endpoints
- idea of session using common control connection
- splitting connection: rate adaptation, disconnection, mobility handling
- general issue: breaks end-to-end semantics: applications might get illusion of successful completion
61
What is PCMP?
Persistent Connection Management Protocol: Session layer over TCP/TLS with 2 proxies
shared session context per user; multiplex transport sessions
persistent application connections: state + buffering at client & proxy
62
How does PCMP work?
1. registration of car at drive-thru client
2. once car connects at wireless network: connect to drive thru proxy
3. Packets get forwarded and responses are sent back
4. once car leaves the hotspot, application responses are buffered (at drive-thru proxy)
5. car reconnects to drive thru client
6. responses get forwarded to client via drive-thru client
