星期六 10 一 2009
Joost: A Measurement Study
Posted by yishuai under Streaming
Yensy James Hall, Patrick Piemonte, and Matt Weyant
这是一篇技术报告。
VoD
1. Meas
- 3 nodes(1 MAC after cable modem, 2 PC after university switch)
- Wireshark
- nineteen days, 64 GB
2. Analysis
- - found geographical location: http://api.hostip.info/get_html.php?ip=
- - could not find the location of 31% of the IP addresses which contributed 4.58% of the data
3. App behavior
- - based on XULRunner, a Mozilla runtime package [11]
- - CoreAVC H.264 codec from CoreCodec [8].
- - signaling traffic is encrypted
- - tracker.ops.theveniceproject.com
- - supernode snode-1.lid.ops.theveniceproject.com
- - negotiates a port number: highly likely that a modified STUN protocol was used, similar to what [4] and [5] observed in their analyses of Skype. UDP or TCP ports?
- - UDP probe (30 bytes) for live detecting
- - TCP and UDP port number 4166, be used in the Now For Friends edition [12]
4. Infrastructure
- - servers on 5 sites, UK London[10], Netherland(Admin node, SN, Web server), Belgium(content servers), Los Angeles[9], California
5. Control it within a network’s boundary
- - port 33333
6. Ctrl and Data traffic
- - [2] use packet sizes to seperate control and data traffic
- Packets under 200 bytes are considered control packets - 45%, and data packets are 1000 bytes or more - 55%.
7. Traffic
- - 700 kbps down and 100-150kbps up
- - rate H.264 : MPEG2 : MPEG4 = 1:2.5:1.75
8. Peer
- - Geographic Locality
- - experiment 1 fails to get a 1min head peer in local subnet
- - is this real? only one experiment
- - if real, need improve here
- - experiment 2, in all 6918 peers in 23hr, 1:1:1 from UK,US,European.
- Is peer distribution related by time zone? it is possible
- - experiment 1 fails to get a 1min head peer in local subnet
- - 6918 peers in 23hr
- is big enough? what is it in PPLive? "relatively low user population"
- - "correlation between geographical distance and amount of data was -0.013"
- - Joost server provides 71.24% of the incoming data
- - How long joost keep download data for sharing? in memory? monitoring the memory.
- - BT Tit-Tac is inappropriate for joost and IPTV/IPVOD.
9. RTT
- - RTT by data size is nonsense. it can not prove RTT is/is not an affect factor for joost peer selection.
- - measure RTT by scriptroute’s ICMP and TCP ping
- - home pc download mainly from Joost server. LAN pc download 52% from Joost server.
- - RTT to joost server is 90-110ms, Hop count: 15-18
- - Have new feature to select nearest joost server
- - roughly 80% of university client peers are within 200ms but only 10% of home client peers are within the same threshold, other > 600ms.
- so home pc always download from joost server
- - meas Hop count by scriptroute’s ICMP and TCP traceroute
- - "it is unlikely that Joost selects peers based on topological locality"
10. fairness:
- - sent/receive = 1/2
- - because server provide at least 2/3.
11. Refer
[3] J. Liang, R. Kumar, and K. W. Ross. The KaZaA Overlay: A Measurement Study. In Computer Networks Journal (Elsevier), 2005.
- A comprehensive measurement study of KaZaA’s overlay structure is undertaken in [3]
- looked in depth at the KaZaA and FastTrack overlay networks with both active probing and passive trace analysis.
- The identification of peer-to-peer traffic
- Techniques to analyze and characterize peer-topeer traffic and file-sharing workloads
- The development of crawling systems for Gnutella, Napster, and KaZaA
- A study of scalability issues in the Gnutella network
- A study of the problem of search and replication strategies in unstructured peer-to-peer networks
- A study about the advantages of designing unstructured peer-to-peer systems based on super peers
[2] K. Gummadi, R. Dunn, S. Saroiu, S. Gribble, H. Levy, and J. Zahorjan. Measurement, Modeling, and Analysis of a Peer-to-Peer File-Sharing Workload. In Proceedings of SOSP, 2003.
- [2] discovered that KaZaA does not take advantage of locality to manage its overlay network and proposed performance improvements if KaZaA were localityaware.
[4] S. Guha, N. Daswani, and R. Jain. An Experimental Study of the Skype Peer-to-Peer VoIP System. In Proceedings of IPTPS, February 2006.
[5] S. Baset and H. Schulzrinne. An Analysis of the Skype Peer-to-Peer Internet Telephony Protocol, http://arxiv.org/abs/cs/0412017v1, 2004.
- [4] and [5] examined Skype’s application behavior and supernode infrastructure.
[6] B. Cheng, X. Liu, Z. Zhang, and H. Jin. A Measurement Study of a Peer-to-Peer Video-on-Demand System. In Proceedings of IPTPS, February 2007.
- [6] studied an on-demand P2P video system called GridCast. GridCast was deployed in China in 2006 to over twenty thousand users, hundreds of which were supported concurrently.
Misc
- [7] V. Paxson. Strategies for Sound Internet Measurement. In Proceedings of ACM SIGCOMM Internet Measurement Conference’04, Taormina, Italy, November 2004.
- [1] S. Ali, A. Mathur, and H. Zhang. Measurement of Commercial Peer-To-Peer Live Video Streaming. In Workshop in Recent Advances in Peer-to-Peer Streaming, August 2006.
- http://developer.mozilla.org/en/docs/XULRunner
- http://www.joostteam.com