1 Optimal Multicast Smoothing of Streaming Video over an Internetwork S. Sen, D. Towsley, Z-L. Zhang, J. Dey Presented by : Shubho Sen
2 Streaming VBR Video Distribution Problem Setting One-many streaming of prerecorded video High bandwidth, multi-timescale burstiness Heterogeneous client and network resources Goal: Efficient transmission schemes Use application-aware multicast of smoothed video Q. how should VBR video be shaped/smoothed for transmission through network? Video Server Client
3 OutlineOutline Review single link smoothing problem Multicast smoothing and differential caching Optimal smoothing for multicast problem Benefits - trace based evaluation Summary
4 Single Link Problem B s bits source buffer at proxy A(t) bits arrive by time t S(t) bits sent by proxy by time t B bits sink buffer at client D(t-w) bits playback by time t w startup delay proxy B D(t-w) A(t) S(t) client streaming video BsBs Do workahead transmission Goal : Reduce peak and variability on path from proxy to client
5 Single Link Smoothing Solution [Salehi:96] O(N) shortest path algorithm finds schedule S that minimizes peak rate Smoothing benefit increases with buffer size Feasible schedule should not overflow or underflow buffers : Lower Constraint = max{ D(t-w), A(t) - B s } Upper Constraint = min{ D(t-w) + B, A(t) } rate variance Upper Lower Cum. bytes w S time t
6 root - 0 b i - buffer size at node i D - consumption schedule for leaf node x S i - transmission schedule for incoming link of node i (link i) Multicast Smoothing Use internal buffers for Differential Caching - Store difference between transmissions to clients with smaller and larger buffers Q. What should the smoothed transmission schedule be along each link ? b1b1 bibi b0b0 b2b2 A SiSi S2S2 S1S1... D D byby bxbx SxSx SySy clients... root Link i Node i
7 Multicast Smoothing Problem Set of schedules {S i } should be feasible S x S i - b i Inter-related set of schedules {S j } Approach : Transform to multiple independent single-link problems A D byby bxbx Node i Parent bibi D bPbP... SiSi SxSx Root b0b0 Node x Link i Link x
8 Transformation to Single Link Problem B s,B i : Effective buffer capacities Can show B s = b 0 + …. + b p B i = ? A D BsBs BiBi SiSi Independent single link problem D byby bxbx Node i Parent bibi D bpbp S i = ? A... Root b0b0 Link i Source buffer Sink buffer
9 Computing B i Example: EB x = b x, EB y = b y EB i = b i + min (EB x, EB y ) B i depends on children: First estimate = EB i EB i = b i + b x bibi D byby bxbx Node i Parent bibi D bpbp Link i bxbx k is child EB m = b m, if m is a leaf = b m + min (EB k ), otherwise
10 Computing B i (contd) B i depends on parent : cannot exceed B p depends on other nodes Recursively compute B i D byby bxbx Node i Parent bibi D bpbp Link i B i = min (EB j | j on path to root) bzbz EB i EB z Link z EB p
11 Multicast Smoothing Solution Converted multicast problem to multiple independent single link problems Compute S i for each problem - use along link i Key Properties Set {S i } is feasible {S i } : set of globally optimal smoothed schedules among all feasible sets S i : smoothest schedule along link i
12 Demonstration of Benefits Setting CBR reservation model Complete 3-ary distribution tree, depth = 4, video stored at root Client buffers in (512 KB, 32 MB) Identical buffer sizes at internal nodes Performance metrics Total bandwidth requirements Sum of bw requirements on path to smallest client 17 min MPEG2 Blues Brothers (peak rate = 44 Mbps, mean = 1.48 Mbps)
13 How much does internal buffering and smoothing help ? Substantial benefits with smoothing Small internal buffering gives substantial benefits Useful to place buffers on path to smallest client
14 SummarySummary Developed optimal smoothing algorithm for multicasting in internetworking environments Integrates smoothing with differential caching Demonstrated smoothing benefits with example Rate constrained problem (paper) find minimum buffer allocation to nodes in distribution tree, and set of optimal smoothed transmission schedules