An Intermediary Smoothing Administration for Variable-Piece Rate Gushing Video.


69 views
Uploaded on:
Description
Re-encode the feature stream at diverse quality at intermediary. Quality debasement; ... Intermediary stores beginning piece of prevalent feature streams. Intermediary begins fulfilling customer solicitation ...
Transcripts
Slide 1

A Proxy Smoothing Service for Variable-Bit-Rate Streaming Video Jennifer Rexford AT&T Labs - Research Florham Park NJ http://www.research.att.com/~jrex Joint work with Subhabrata Sen, Don Towsley, and Andrea Basso

Slide 2

Outline Background and inspiration Burstiness of packed video streams Smoothing procedures for put away video Online smoothing of variable-piece rate video Sliding-window smoothing calculation Performance assessment on MPEG follows Integration of smoothing with prefix storing Caching beginning casings of famous video streams Resource assignment over different streams Prototype intermediary smoothing administration Software configuration of intermediary administration in Windows NT MPEG-2 PC-based video spilling testbed Conclusions and continuous work

Slide 3

Video Streaming Applications Live, intuitive Video remotely coordinating, video telephones, and so on. Tight defer limitations to bolster intuitiveness Stored, non-intelligent video Movies, separation learning, Web recordings, and so on. Video recorded ahead of time; free postpone imperatives Live, non-intuitive video Course addresses, news, donning occasions, gatherings Video not recorded ahead of time; free defer imperatives

Slide 4

Network Environment

Slide 5

Challenges of Video Streaming High transfer speed necessities of compacted video 4-6 Megabits/second for top notch MPEG2 streams Burstiness of edge sizes on a few time scales MPEG gathering of-pictures structure (I, P, B outlines) Differences in real life and point of interest inside/crosswise over scenes Bandwidth confinements on customers and connections 10 or 100 Mbps shared neighborhood 27 Mbps link channel, 1.5 Mbps ADSL Lack of end-to-end control of way from source Poor deferral, throughput, and misfortune in the Internet

Slide 6

Compressed Video Streams

Slide 7

Approaches to Handling Variability Constant-piece rate encoding of every stream Adjust nature of encoding to stay at steady rate Quality corruption amid scenes with activity & subtle element Statistical multiplexing of variable rate streams Rely on blending to diminish the total pinnacle rate Limited viability on access joins Selective dispose of parcels/casings in stream Discard bundles/outlines amid transient clog Noticeable debasement in video quality Transcoding or layered encoding to decrease bit rate Re-encode the video stream at various quality at intermediary Quality corruption; hard to transcode at connection speeds

Slide 8

Smoothing Stored Video For prerecorded video streams: All video outlines put away ahead of time at server Prior information of all casing sizes ( f i , i=1,2,..,n) Prior learning of customer cradle size ( b ) workahead transmission into customer cushion 2 1 b bytes n Client Server

Slide 9

Smoothing Constraints Given edge sizes { f i } and cushion size b Buffer undercurrent requirement (L k = f 1 + f 2 + … + f k ) Buffer flood limitation (U k = min(L k + b, L n )) Find a calendar S k between the limitations O(n) calculation minimizes pinnacle and variability U number of bytes rate changes S L time (in edges)

Slide 10

Reducing the Peak Rate

Slide 11

Limitations of Smoothing Model Assumes prerecorded put away video yet… need to bolster live and precorded video Assumes smoothing is performed by server however… server is in the area of another supplier Assumes end-to-end control of the system yet… the Internet is decentralized Assumes server knows the customer support measure yet… the customer might be in an alternate space

Slide 12

Online Smoothing Source or intermediary can defer the stream by w time units: Larger window w lessens burstiness, yet… Larger support at the source/intermediary Larger preparing burden to figure plan Larger playback delay at the customer stream with postponement w gushing video b bytes Client Source/Proxy

Slide 13

intermediary customer An i S i D i-w b Online Smoothing Model Arrival of An i bits to intermediary by time i in edges Smoothing support of B bits at intermediary Smoothing window (playout deferral) of w edges Playout of D i-w bits by customer by time i Playout cradle of b bits at customer transmission of S i bits as a substitute by time i

Slide 14

Online Smoothing Must send enough to maintain a strategic distance from sub-current at customer S i should be at any rate D i-w Cannot send more than the customer can store S i should be at most D i-w + b Cannot send more than the information that has arrived S i should be at most An i Must send enough to evade flood at intermediary S i should be no less than An i - B max{D i-w , An i - B} <= S i <= min{D i-w + b, An i }

Slide 15

Online Smoothing Constraints Source/intermediary has w outlines in front of current time t: don\'t have a clue about the future number of bytes U L ? time (in casings) t t+w-1 Modified smoothing requirements as more edges arrive...

Slide 16

Smoothing Star Wars GOP midpoints 30-second window 2-second window MPEG-1 Star Wars,12-outline gathering of-pictures Max outline 23160 bytes, mean casing 1950 bytes Client support b=512 kbytes

Slide 17

Reducing Computational Complexity No compelling reason to figure plan at each time unit Limited data from new casing landings Limited effect on direction of the timetable Execute online calculation each a period units Perform O(w) work each a period units Limit number of rate changes Performance suggestions Very little increments in pinnacle and difference of rates Setting a = w/2 performs just about and a = 1

Slide 18

Parameters in Smoothing Model Algorithm parameters Window w (in number of casing spaces) Client cushion size b (in bytes) Source/intermediary cradle size B (in bytes) Computation interim an (in edges) Frame-size expectation interim p (in edges) Performance measurements Peak rate of the smoothed stream Coefficient of variety (standard-deviation/mean) Effective transfer speed (given support and misfortune rate)

Slide 19

Peak Rate versus Window Size (changing customer support size for MPEG-1 Wizard of Oz ) Dramatic reduction in transfer speed variability Online calculation approaches disconnected plan Ten-second window gives the vast majority of the increase

Slide 20

Peak Rate versus Customer Buffer (shifting window size for MPEG-1 Wizard of Oz ) Significant decreases with a couple Mbytes of support Diminishing returns for bigger customer cushion sizes Window size w ought to scale with cradle size b

Slide 21

Proxy versus Customer Buffer (fluctuating forecast under 512-kbyte all out cradle & 30-outline window) Need cushion at every end for good execution Even support for vast P, more at intermediary for little P Simple expectation plans are extremely successful

Slide 22

Prefix Caching to Avoid Start-Up Delay Avoid start-up deferral for prerecorded streams Proxy stores beginning a portion of famous video streams Proxy begins fulfilling customer ask for all the more rapidly Proxy asks for rest of the stream from server smooth over substantial window immediately Use prefix reserving to stow away other Internet delays TCP association from program to server TCP association from player to server Dejitter cushion at the customer to endure jitter Retransmission of lost bundles apply to "point-and-snap" Web video streams

Slide 23

New Questions Video spilling convention How to get the intermediary in the way? How to get an underlying duplicate of the prefix? How to recover the rest of the casings of the video? Smoothing model What changes in the smoothing limitations? What changes in the fundamental execution properties? Intermediary asset portion How much prefix is expected to conceal Internet delays? How to allot amongst reserving and smoothing? How to distribute assets over numerous streams?

Slide 24

Protocol Issues Ensuring that solicitations experience the intermediary Configuration of intermediary in customer program or player Placement of straightforward intermediary in the way Caching of the underlying casings of the video Server replication of the prefix Proxy prefetching of the prefix Proxy reserving of prefix after first demand Transparent recovery of residual edges Range ask for operation in HTTP 1.1 Absolute situating in RTSP

Slide 25

Changes to Smoothing Model Separate parameter s for customer start-up deferral Prefix store stores the principal w-s outlines Arrival vector An i incorporates stored outlines Prefix support does not void after transmission Send whole prefix before flood of b s Frame sizes might be known ahead of time (stored) An i b S i D i-s b c b p

Slide 26

Performance Evaluation Comparison to unique internet smoothing model Pro: can have extensive window and little start-up postponement Pro: execution is for all intents and purposes indistinct Con: putting away prefix almost duplicates cradle necessity Con: might be hard to smooth at start of video Allocation of prefix and smoothing cushions Small prefix cushion limits size of smoothing window little window w confines workahead smoothing Large prefix cradle limits size of smoothing cushion little b s requires forceful transmission plan

Slide 27

Peak Rate versus Window Size (changing aggregate intermediary cushion size for MPEG-1 Wizard of Oz ) Convex, container formed bend of pinnacle rate versus support Simple double scan for ideal portion Heuristic: pick biggest w that does not oblige b s

Slide 28

Peak Rate versus Prefix Buffer Size (shifting aggregate intermediary cushion size for MPEG-1 Wizard of Oz )

Slide 29

Allocating Resources Across Streams Performance issues Limited cradle ( M ) and/or data transmission ( B ) at intermediary Collection of V recordings with various ubiquity Videos with various successions of edge sizes Optimization issue Allocate prefix support b p for every video v =1,… , V Allocate smoothing cushion b s for each of n v demands Obey imperative on cradle ( M ) or transfer speed ( B ) Minimize the use of the other asset ( M or B )

Slide 30

Simplifying the Problem Complex asset designation issue Assign b p , b s , and w for every video v Buffer prerequisite: total v {b p (v) + n v * b s (v)} Bandwidth necessity: total v {n v * peak(v)} Reduce issue to selecting w for every video Select same b s and w over all solicitations for v Select prefix cradle b p as first w-s outlines Sel

Recommended
View more...