Presentation is loading. Please wait.

Presentation is loading. Please wait.

A Hierarchical Characterization of a Live Streaming Media Workload E. Veloso, V. Almeida W. Meira, A. Bestavros, S. Jin Proceedings of Internet Measurement.

Similar presentations


Presentation on theme: "A Hierarchical Characterization of a Live Streaming Media Workload E. Veloso, V. Almeida W. Meira, A. Bestavros, S. Jin Proceedings of Internet Measurement."— Presentation transcript:

1 A Hierarchical Characterization of a Live Streaming Media Workload E. Veloso, V. Almeida W. Meira, A. Bestavros, S. Jin Proceedings of Internet Measurement Workshop, ACM SIGCOMM, Nov. 2002

2 2 Outline Introduction Data Collection Client Layer Characteristics Session Layer Characteristics Transport Layer Characteristics Conclusions

3 3 Introduction Workload characterization is important for Performance evaluation and prediction Capacity planning Rejecting client for a live stream is not a viable solution Value of live streams is the liveness Lose paying customers

4 4 Introduction Only a small number of studies on characterizing pre-recorded streaming media workloads This paper provides a characterization for live streams

5 5 Introduction Compare to Stored streams, Live streams exhibit Stronger temporal patterns of workload Fewer possible VCR functions Less correlations between different variables Users are less likely to stop viewing when QoS degrades

6 6 Data Collection A popular live show in Brazil “Reality TV Show” in early 2002, last for 90 days The live streams provided feeds captured from one of the cameras embedded in the environment surrounding the contestants

7 7 Data Collection For each entry of the log, it contains Client identification—e.g., IP address, player ID Client environment specification—e.g., OS version, CPU Requested object identification—e.g., URI of stream Transfer statistics—e.g., loss rate, average bandwidth Server load statistics—e.g., server CPU utilization Other information—e.g., referer URI, HTTP status Timestamp in seconds of when log entry was generated.

8 8 Data Collection

9 9 Client Layer Characteristics Focus on the characteristics of the client population Clients are identified by the unique player ID

10 10 Client Topological and Geographical Distribution Follow a Zipf profile with parameter α=1.29, 1.49 and 5.4 respectively

11 11 Temporal behavior of number of active clients Diurnal Effect on the live content Periodic Depends on the day of week

12 12 Client Arrival Process Client arrival process is not poisson Can be estimated by a sequence of piece-wise-stationary Poisson arrival processes Interarrival time of clients from logs Interarrival time of a piece-wise-stationary Poission process

13 13 Client Interest Profile Using transfer frequency as a measure of client interest in the content Client interest in a single object follows a Zipf distirbution

14 14 Session Layer Characteristics Focus on individual client activity The trace does not explicitly identify the delimiters of a session The authors choose a session timeout parameter T off to determine the number of sessions T off = 3600 seconds

15 15 Session ON/OFF Time ON times are highly variable Due to live content instead of temporal behaviors Lognormal OFF times form ripples around specific values In multiple of days => revisting daily or every two days Exponential Session ON Time vs Session starting time

16 16 Transport Layer Characteristics Focus on individual unicast data transfers Temporal behavior of no. of concurrent transfers Periodic over a weekly and daily period Similar to the temporal behavior of no. of active clients

17 17 Temporal behavior of transfer interarrival times Request arrival process is also periodic and non-stationary Due to the diurnal behavior

18 18 Transfer Length & Client Stickiness Similar to the session ON time The long tail shows the willingness of the client to “stick” to the live object

19 19 Transfer Bandwidth Bounded by client connection speed Bounded by congestion

20 20 Representativeness of Findings Compared the findings with another live show “Live News & Sports” Sport news & soccer players interviews 28558 requests from 12867 distinct clients within 2 weeks interarrival times depends on the content

21 21 Conclusions Client Layer Arrival process can be modeled by a piece-wise stationary Poisson process Identity of the client making a request can be modeled by a Zipf distribution Session Layer ON times follows Lognormal distribution OFF times follows exponential distribution Transfer Layer Arrival process can be modeled by a piece-wise stationary Poisson process Transfer bandwidth is primarily determined by client connection speed while 10% of transfers are being severely limited by congestion

22 22 Comments Piece-wise Poisson Process A good way to model the client arrival process But we need a priori knowledge of the average client arrival rate with a number of short period The client arrival pattern also depends on the content Hard to be used


Download ppt "A Hierarchical Characterization of a Live Streaming Media Workload E. Veloso, V. Almeida W. Meira, A. Bestavros, S. Jin Proceedings of Internet Measurement."

Similar presentations


Ads by Google