Multimedia Networks Flashcards
What are the 3 application types of multimedia networking?
- streaming stored
audio/video - Conversational voice/video
over IP - streaming live audio/video
What is client-side buffering?
- initial fill of buffer
- playout begins (at t_p)
- buffer fill level varies over
time (fill rate x(t) varies and
playout rate r is constant)
Why is HTTP used for streaming multimedia?
- file retrieved via HTTP GET
- send at max rate under TCP
- fill rate fluctuates due to TCP
congestion - larger playout delay = smooth
TCP delivery rate
Why is UDP used for streaming multimedia?
- server sends at rate
appropriate for client - short payout delay to remove
network jitter - error recovery at application
level
What is the equation for playout time for first packet in talk spurt?
playout time_i = t_i +d_i + Kdv_i
how does a one size fits all service model work?
increase link capacity so congestion doesn’t occur
what are the disadvantages of a one-size fits all service model?
- low complexity of network
mechanisms - high deployment costs
What is a multiple classes of service model?
- partition traffic into classes
- network treats different
classes differently
What are the 4 principle QoS guarantees for network mechanisms
Service Level Agreement
Policing mechanism
Unused bandwidth
Call admission
Why do we need an SLA (service level agreement)?
packet classification = router treat different traffic classes accordingly
packet marking = router can distinguish packets belong to different classes
Why do we need a policing mechanism?
- provide degree of traffic isolation among classes
- one class not adversely affected by another that’s misbehaving
Why do we need an unused bandwidth mechanism?
- whilst providing isolation between flows, want to use resources efficiently
- flows allow use of another flow’s unused bandwidth at any time
Why do we need call admission?
- sufficient resources not always available
- before accepting new call, flows declare QoS requirements
- network can deny if cannot meet needs
What is priority queuing?
multiple queues to classify as high/low priority
What is weighted fair queueing?
- generalized round robin
- each class gets weighted amount of service in each cycle
What are the tasks of queue management?
- remove packets from a queue on request from packet scheduler
- discard/remark packets if queue full/saturated
- proactively avoid queue becoming congested
What is packet removal/discard policy and what are some examples?
if packet arrives to full queue what do we do?
- tail drop = drop arriving packet
- priority = drop on priority basis
- random = drop randomly
What is random early detection (RED)?
- drop packets randomly from selected flows
- when queue length exceeds threshold
What is a 2 queue length threshold?
- if buffer average below min threshold, no congestion
- if buffer above max threshold, drop packets with p= 1
- if buffer between min and max, drop packets with p = p
What is random early detection with in/out (RIO)?
- assumes edge router marking of packets conforming to SLA
- conforming packets in profile
- non-conforming out of profile
- out of profile packets dropped first if core router has congestion
What is the goal of a policing mechanism?
shape traffic so it doesn’t exceed declared/agreed parameters
What are the 3 commonly used criteria for policing mechanisms?
- Average rate: packets sent per unit time
- Peak rate
- Burst size: max no of packets sent consecutively
What do combine token/leaky bucket and weighted fair queueing provide?
Upper bound on delay
What are the 2 approaches on internet quality of service?
Fina grained approach
- provide QoS for individual applications or flows
Coarse grained approach
- provide QoS to large classes of data or aggregate traffic
What is resource reservation protocol (RSVP)?
network-level signalling protocol operating on top of IP
How does RSVP work?
- host requests specific QoS for particular flow
- routers provide requested QoS along paths
What are RSVP routers responsible for?e
- dedicate requested resources
- maintain soft state for signalled connection
- accept/refuse new connections based on available resources
- guarantee service level agreement of existing calls
How does the sender work is RSVP?
send message containing TSpec to receiver
What is TSpec?
flow traffic characteristics
How does the receiver work in RSVP?
- looks at sender traffic specifications TSpec
- needs to know path of packet
- initiate establishment of resource reservations at each router on the path
Where are reservations sent in RSVP?
back to sender
- if reservation can be made, RESV message request to next rout
What is an RSEV message?
message containing RSpec
What is RSpec?
service request from networks
What are the pros/cons of RSVP?
pro:
- simple to increase/decrease
level of resource allocation
con:
- required for each flow state
info to be stored on
particular route
- each reservation needs
memory
- routers have to classify,
police and queue each flow
- need admission control decisions
What is call admission control?
looks at TSpec and RSpec and decides per flow if service can be provided
What is resource reservation for RSVP?
bandwidth and buffers reserved for data flow
How does traffic policing work in RSVP?
per-packet decision to ensure all flows conform to TSpec’s
What is the differentiated services model?
- simple and more scalable
- per flow service replaced with per aggregate service
- complex processing moved form core to edge of network
- allocates resources to small number of traffic classes
What is the Differentiated Service (DS) field used for?
identify per hop behaviour at each node
What is required for DS?
- service level agreement
What does a service level agreement (SLA) include?
- traffic conditioning agreement (TCA)
What is a traffic conditioning agreement (TCA)?
defines nature of various traffic classes and services
How does DS work for the edge router?
- per-flow (aggregate) traffic management
- marks packets as in-profile and out of profile
How does DS work for the core router?
- per class traffic management
- queueing and scheduling based on marking at edge
- preference to in-profile packets
What are the 2 ways of ensuring traffic conformance for TCA at the edge router?
Traffic classification
Traffic conditioner
What is traffic classification in regards to TCA?
direct packets into various output based on content on packet headers
What is traffic conditioning in regards to TCA?
metering, marking, shaping and dropping
What is packet marking?
- packet marking at edge based on per-flow profile
What are the 2 types of packet marking and how do they work?
class-based marking: packets of different classes marked differently
intra-class marking: conforming (in-profile) portion of flow marked differently than non-conforming
How does forwarding per-hop behaviour work?
- individual routers class differentiation are called PHB
- PHS results in different measurable forwarding performance behaviour
- doesn’t specify what mechanisms to use to ensure required PHB performance
What is expedite forwarding PHB?
low loss
low latency
low jitter
assured bandwidth
end to end services
What is assured forwarding PHB?
delivers aggregate traffic with high assurance as long as traffic doesn’t exceed traffic profile
- not intended for low latency/low jitter
What is expedite forwarding PHB limited by?
link speed of router
What is expedite forwarding PHB implemented by?
using priority/weighted fair queueing
How does assured forwarding PHB work with queue management?
AF has root in RIO queue management
What is multiprotocol label switching (MPLS)?
- viewed by internet as logical link connecting IP routers
- integrates layer 2 switching with layer 3 routing
- separation of control plane and forwarding plane
What does the forwarding plane do?
carries info
What does the control plane do?
set up routes
What so MPLS switching routers do?
- improve forwarding performance using simplified lookup process
- improve scalability using label stacking and merging
- provide traffic engineering via efficient explicit routing
What are some remarks on MPLS?
- doesn’t define another QoS architecture
- uses DiffServ architecture
- processing of aggregate traffic only
- DSCP field not available to MPLS LSR
- mapping needed to map DSCP field into EXP field
