Transcript Document

FEC-Integrated Network Traffic
Shaping Using the NIProxy
Maarten Wijnants, Wim Lamotte
Hasselt University – Expertise Centre for Digital Media (EDM)
Wetenschapspark 2, BE-3590 Diepenbeek, Belgium
{maarten.wijnants,wim.lamotte}@uhasselt.be
Outline
• Background and Motivation
– Error Correction Techniques
• Network Intelligence Proxy
– Objectives & Methodology
• FEC Integration in NIProxy
• Evaluation
– Experiment Description
– Experimental Results & Findings
• Conclusions
15/10/2009
EMERGING2009
2
Background and Motivation
• Exchanging data over computer networks
can lead to corruption
– Data becomes (partly) unusable for receiver
• Data corruption can be caused by
– The loss of entire packets
• E.g. insufficiently capacitated network infrastructure
– The introduction of bit errors
• E.g. signal interference and noise on the channel
• Irrespective of its cause, data corruption is
likely to degrade user experience
– Effort should be made to minimize it!
15/10/2009
EMERGING2009
3
Error Correction Techniques
• 2 data corruption countermeasure categories
– Retransmission-based techniques : Receiver
requests source to retransmit missing or
corrupted data
– Forward Error Correction (FEC) : Sender
supplements source data with redundant info
which allows receiver to repair, to a certain
extent, errors introduced during transmission
• FEC schemes enable lost or damaged data
recovery without incurring RTT overhead
introduced in retransmission-based solutions
15/10/2009
EMERGING2009
4
Error Correction Techniques
• Example FEC scheme: XOR-Based Parity
Coding
– Input = Group of n media packets
– Output = Single parity packet
• Constructed by applying the XOR operator on the bits
stored at identical locations in the n input packets
– At decoding side, parity packet can be used to
recover a singly lost/corrupted packet
• By XOR-ing the (n - 1) correctly received media
packets with the (also perfectly received) parity pack
– Important advantage: Run-time adaptability:
Trade off protection for BW (by changing n)
15/10/2009
EMERGING2009
5
Error Correction Techniques
• Retransmission- and FEC-based schemes
share a common disadvantage
– Both introduce overhead in terms of the amount
of data that needs to be transmitted
• I.e. the BW requirements of data flows are raised
• The surprising scenario might occur where
the addition of error protection yields an
increased instead of a decreased error rate
• Deliberate decision making regarding the
amount of protection to add to network
traffic is advocated!
15/10/2009
EMERGING2009
6
Network Intelligence Proxy
• Network intermediary (a “proxy”)
– Can be incorporated in existing IP networks
• Goal = Optimize QoE of users of
distributed applications
• Approach = Gather context and improve
MM handling capabilities of transportation
network to enable user QoE optimization
– Network traffic shaping
– Multimedia service provisioning
• NOT transparent
15/10/2009
EMERGING2009
7
Network Intelligence Proxy
Methodology
• NIProxy introduces “intelligence” in the
networking infrastructure
– 2 distinct sources of contextual info are queried
• Source 1: Transportation network
– Contextual knowledge = Quantitative network-related
measurements and statistics
– Obtained through active network probing & monitoring
• Source 2: Distributed application
– Contextual knowledge = Any application-related
knowledge that is deemed relevant
– Needs to be provided by the application software
15/10/2009
EMERGING2009
8
Network Intelligence Proxy
Network Traffic Shaping
• Orchestrate bandwidth consumption by
arranging flows in a stream hierarchy
– Tree-like structure; expresses flow relationship
– Internal nodes implement bandwidth
distribution strategy
• Mutex : Available bandwidth BW allotted to child with
largest still satisfiable bandwidth requirement
• Percentage : Each child i is granted its
corresponding percentage value
of the
distributable bandwidth BW, i.e.
– Leaf nodes correspond with actual flows
• Discrete leaf : Switch BW usage of associated flow
between discrete number of levels
15/10/2009
EMERGING2009
9
Network Intelligence Proxy
Network Traffic Shaping
• Sibling dependencies framework
– Enables dependencies to be enforced between
sibling nodes in the stream hierarchy
– Currently only 1 type of dependency defined,
namely SD_BW_ALLOC_CONSTRAINED
• Set of supported sibling dependency types readily
extensible
– SD_BW_ALLOC_CONSTRAINED dependency
between sibling nodes A and B specifies that B
is allowed to consume bandwidth if and only if
A’s bandwidth consumption is non-zero
• Node A can “borrow” bandwidth assigned to B
15/10/2009
EMERGING2009
10
Network Intelligence Proxy
Multimedia Service Provision
• NIProxy acts as service provision platform
– In-network execution of (context-aware)
services on transported data
• Implemented using a plug-in based design
– Each service corresponds to a NIProxy plug-in
• Service cooperation through chaining
• NTS and MM service provision integrated
in an interoperable manner!
– Services can query/influence the bandwidth
distribution strategy devised for clients
– Unlocks extra QoE optimization possibilities
15/10/2009
EMERGING2009
11
FEC Integration in NIProxy
• Given its negative impact on user experience,
techniques to counter lost or damaged data
are meaningful extensions of the NIProxy’s
feature list
• Adaptive XOR-Based Parity coding
implemented as NIProxy service
• Integrated approach with NIProxy NTS
– FEC-generated network traffic might consume
significant amounts of bandwidth
• Should be reckoned with by NIProxy’s NTS mechanism
• Necessitates FEC traffic inclusion in stream hierarchy
15/10/2009
EMERGING2009
12
FEC Integration in NIProxy
• FEC incorporation in stream hierarchy
– Redundant FEC parity data is represented as discrete stream
hierarchy leaf node
• Defines a discrete bandwidth consumption level for each
supported input packet grouping size
– FEC data also needs to be adequately related to the media
stream it protects (JSCC)
• Deliberately amortize BW that has been reserved for FECprotected traffic among the media data and its FEC overhead
• In this paper: By using a Percentage node
– Adjusting the percentage values assigned to both nodes allows the
JSCC process to be controlled
– SD_BW_ALLOC_CONSTRAINED dependency between the
nodes representing the media and its FEC protection
• FEC can consume BW if and only if associated media flow is enabled
15/10/2009
EMERGING2009
13
FEC Integration in NIProxy
• Operation of the NIProxy FEC service
– Performs 2 initialization tasks on discovery of
network stream eligible for FEC protection:
• Instantiate a XOR-based parity encoder
• Inform NTS process of possibility to FEC protect the
stream and the thereby associated BW requirements
– Main processing loop:
• Service exploits its interface with NTS to determine
discrete level to which the FEC data for the media flow
that is being processed is currently set
• FEC encoder is switched to the input grouping size that
is associated with this level
• Packet is fed encoder (possibly producing parity packet)
15/10/2009
EMERGING2009
14
Evaluation
Experimental Setup
• FEC support advantageously influences
NIProxy’s user QoE optimization capabilities?
• Video streaming case study
High capacity; Error-free
15/10/2009
Resource constrained; Error-prone
EMERGING2009
15
Evaluation
Experiment Description
• MM server maintained 2 simultaneous RTP
video sessions with client: VS1 and VS2
– Video data emitted in unprotected form
– NIProxy had its FEC service loaded
• Parity coding per 3 or per 6 input packets
– Only video session VS2 was marked as being
eligible for receiving FEC protection
– Identical video fragment streamed over both
sessions to allow meaningful comparison
– NIProxy video transcoding service also loaded
• To address bandwidth shortage on the access network
15/10/2009
EMERGING2009
16
Evaluation
Experiment Description
• Experiment was executed twice
– Once without and once with the netem
component introducing packet loss on last mile
– Access network throughput artificially modified
at predefined points in time (5 times in total)
• Investigate effect on the way the NIProxy shaped the
network traffic destined for the receiving client
– All other conditions remained constant
• Bandwidth modifications conceptually divided the
experiment into 6 discrete intervals
15/10/2009
EMERGING2009
17
Evaluation
Experiment Description
• Stream hierarchy which steered the
shaping of the network traffic
Split access bandwidth
equitably
FEC protected
Video Session VS2
Static JSCC
(90%-10%)
XOR disabled
n = 3; n = 6
Unprotected
Video Session VS1
15/10/2009
EMERGING2009
SD_BW_ALLOC_
CONSTRAINED
18
Evaluation
Experimental Results
• Execution 1: Error-free environment
All streams at max quality
FEC coding disabled
VS2 transcoded to lower quality
Access BW gradually more constrained
Increasing flow BW reductions required
15/10/2009
EMERGING2009
19
Evaluation
Experimental Results
• Execution 2: 10% packet loss
Playback VS2 less distorted! 
15/10/2009
EMERGING2009
Video playback at
 destination no longer
perfect!
Residual vs original packet
loss VS2 = 92 vs 210
20
Evaluation
Experimental Results
• Findings and observations
– Capacity of client’s access connection respected
– Delineated BW distribution strategy successfully enforced
• Access bandwidth shared equitably among video sessions
– Example of potential of supporting interoperation between
NIProxy services and bandwidth brokering
• E.g. JSCC process steered entirely by NIProxy’s NTS
– JSCC might require quality of MM data to be reduced to
accommodate its FEC protection
• Therefore quality VS2 sometimes lower than VS1
• FEC overhead however enables packet loss recovery
– Playback VS2 smoother and less perceptually degraded
– Lower quality yet less distorted = More enjoyable viewing
experience than high-quality distorted video (subjective)
15/10/2009
EMERGING2009
21
Conclusions
• MM data might arrive in corrupted form during its
propagation through error-prone networks
– Typical outcome = Deteriorated media presentation
– Likely source for user frustration
• FEC schemes possess the ability to alleviate
detrimental effects of data corruption
– Enable receivers to repair compromised data
• FEC incorporated in NIProxy (XOR parity code)
– FEC operations directed by NTS  Ensure XOR BW justified
– Evaluated using video streaming case study
– Results corroborate that FEC coding is valuable addition to
NIProxy’s toolset to improve user QoE
15/10/2009
EMERGING2009
22