Week 5 - Advanced Routing and QoS Flashcards
Define multicast?
The transmission of information to more than one destination.
What is IGMP?
One of the three main IP protocols (other htan ICMP and IP) - Internal Group Management Protocol (IGMP) is used for multicast messaging. When a computer wants to join a group it sends an IGMP JOIN message to its router. Routers then periodically poll hosts in the network to ask if they are still in teh group. Once accepted, the router joins the multicast tree.
What is DVMRP? What is MOSPF?
Distance Vector Multicast Routing Protocol (DVMRP).
- All routers receive the traffic for all groups.
- Router without any members on a group send Prune message up the tree.
- Router without any neighbours as members remove multicast entries.
- Process is repeated periodically.
Very inefficient especially for small groups.
MOSPF is a multicast extensio to OSPF. In OSPF we already know all the node route information, MOSPF proposes to use that to multicast the packages. This also has a big disadvantage of each router having to know all the group memberships.
What is PIM-SM?
Protocol Independent Multicast - Sparse Mode
DVMRP works the way it does because the router does not know what to do to join a group.
PIM-SM each group has a Rendez Vous Point which needs to be known in advance (provided outside the system). The router then sends a unicast message to the RP to join. Hosts wishing to send to the multicast group just send to the RP.
Note: Only specific sources can send to RP (this is to avoid DDOS attacks).
What challenges remain for Multicast?
Technical:
- Each group requries state (routing tables) in every router in the path.
- Difficult to be used for time shifted content
- Difficult to achieve reliability and congestion control
Commercial:
- ISPs like to sell bandwidth. Multicast saves bandwidth so is bad for business.
Availability:
- Multicast is only available in limited intra-domain deployments
What is SDN? And the pros/cons?
Pros:
- Allows for much more elaborate routing.
- Centralization allows for decisions to be made based on many packet fields, simplifies the router control plane
Cons:
- Resilience becomes more problematic
- Memory in the forwarding plane increases
How does the forwarding table work?
Packets that don’t match any rule may be escalated to the controller. Controller then calculates rule and send back to the router.
What is the most common arrival distribution: Poisson?
What is the most common servicing disribution?
What is Kendals notation?
Population size = total number of packets
queue discipline = FIFO example
M/M/1 = Markovian/Markovian/1 server is the most common network format
Or M/D/1 = Markovian/Deterministic/1 if service is a fixed rate
What assumptions do we make when doing queueing calculations?
- FIFO
- No bulking or reneging
- Arrivals are indepdendent (markovian)
- Service times are independent
- Arrival and service rates remain stable
What are the formulas for queueing calculations for average number of packets in the queue / average time packets spends on the queue?
Often each output port has many queues (one per flow, one per class), how do we decide which packet to choose next? Pros and cons of each approach?
Fair Queueing (FQ) - Round Robin. Affects greedy flows, also benefits flows with large packets by default.
Bit Round Fair Queueing (BRFQ) - Emulates a bit-by-bit round-roin discipline but whole packets are forwarded in each round. This achieves fair queueing independently of the packet sizes. Good for classless best-effort traffic only
Weighted Fair Queueing (WFQ) - Round robin with pre-assigned weights. Weights may be applied based on flow resources for example (i.e. flow with 512Kb/sec gets more than the 64Kb/sec resource).
Priority Queueing (PQ) - queues are assigned priorities and always served in strict order of priority, with priority queue always served first. Good for important traffic but can lead to starvation. This is resolved with Rate Limits. A priority queue is served until their rate limit (bandwidth rate cap) then other queues are served the same way.
How does weighted fair queueing work in practice?
Considering three queues A,B,C with packet sizes 200, 50, 100, how are the queues served? Assuming now that the weights associated with the three queues are 1,2,3, how are the queues served?
What is Random Early Detection (RED)?
- Calculate the average queue size (TH=threshold)
- If avg < THmin then queue packet
else if THmin<=avg<THmax then calculate probability to discard packet, if within probability discard otherwise queue
finally if THmax<=avg then discard packet
How is Quality of Service (QoS) maintained in a network?
1.Resource reservation
2.Sensitive scheduling
What is the issue with RSVP in relation to QoS? And whats the alternative?
For every connection, a reserve has to be stored in every router on the path. This has a big cost in the core routers in the Internet. Instead of every router holding state over the route for each channel, edge routers are setup with per class traffic management. Buffering and scheduling is based on marking at the edge.
What are the Per-Hop behaviours (PHBs)?
