Presentation on theme: "WHITEBOARD: A DISTRIBUTED APPLICATION OVER XCAST WHITEBOARD: A DISTRIBUTED APPLICATION OVER XCAST Justinus Andjarwirawan Justinus Andjarwirawan."— Presentation transcript:
WHITEBOARD: A DISTRIBUTED APPLICATION OVER XCAST WHITEBOARD: A DISTRIBUTED APPLICATION OVER XCAST Justinus Andjarwirawan email@example.com Justinus Andjarwirawan firstname.lastname@example.org Committee: Prof. Kanchana Kanchanasut Dr. Yasuo Tsuchimoto Dr. Sumanta Guha Committee: Prof. Kanchana Kanchanasut Dr. Yasuo Tsuchimoto Dr. Sumanta Guha AI3 research presentation - 2006.04.20
2 Contents 1.Introduction Whiteboard overview and background Problem statement 2.Underlying Protocols 3.Implementation 4.Evaluation 5.Conclusion
3 Whiteboard: a computer software application that simulates actual whiteboards, allowing one or more people to write or draw images on a simulated canvas. This is a common feature of many virtual meeting, collaboration, and instant messaging applications. (from: wikipedia) 1. Introduction
5 Methods used by existing Whiteboards The two methods widely used by those existing whiteboards: Unicast: information is sent from one point to another point. A whiteboard having many destinations means sending multiple similar packets! Multicast: information is sent from one point to a set of other points (one packet sent to a group of destinations at one time), but with the need of additional routing protocols (an overhead).
6 Problem Statement Whiteboards normally involved a small number of users, hence Multicast will be wasteful. What to do with this problem: Develop a whiteboard over an alternative protocol
7 What Network Protocol to Use? Based on the problem statement: Unicast is wasteful for numbers of users because it is sending similar packets to multiple destinations. Multicast is wasteful for small groups communication because of the routing mechanism. Another protocol that may solve this? XCAST
8 XCAST XCAST (eXplicit Unicast): a multi-destination delivery system in the network. Information is sent from one point to a set of other points, with multiple destination addresses in each single packet. (a regular IP packet contains only one source address and one destination address)
9 Why XCAST? 1.XCAST works on routers with existing Unicast routing information. There is no need for additional routing protocol like Multicast which can lead to a delivery time overhead. 2.Participants are unique and identified clearly by individual IP addresses, therefore individual access lists, policies and accounting can be implemented.
10 But … XCAST has some disadvantages and limitations: Overhead in packet header size, because of the multi- destination entries More header processing on the fly Limited number of destinations (logically maximum 9 for IPv4, 127 for IPv6) XCAST packets cannot be fragmented into smaller MTUs (Maximum Transmission Units).
11 XCAST6 Limitations XCAST packets cannot be fragmented, meaning each packet must not exceed MTU (Maximum Transmission Unit) size. MTU = the largest possible packet size that a network can transmit at a time (in this case Ethernet = 1500 bytes). If exceeds then MTU on the network will divide into smaller packets. But this is not possible for XCAST because of its header structure.
12 XCAST Packet headerdest. add.data MTU 1500 bytes headerdest. add.data The more # of destinations the less data space 40 bytes
13 What to do with MTU limit Total size of maximum 1500 includes: XCAST packet header, destination addresses in the extension header and data payload. Longer data payload is split into smaller chunks at the application level implementation, before forming to packets. Small chunks size must also reserve space for destination addresses.
14 2. Implementation The latest version of MBONEs Multicast whiteboard program has many bugs and the libraries used in the source code are outdated. Originally works on IPv4. The implementation of XCAST Whiteboard still follow the MBONEs whiteboard mechanisms and routines. Development under C language. Operating system: Linux
15 The Programs Flow canvas drawing incoming draw commands from remote users incoming draw commands from remote users outgoing draw commands to remote users outgoing draw commands to remote users receiving socket (listening) receiving socket (listening) sending socket
16 Combining Tcl/Tk and C Tcl/Tk canvas code (Tcl/Tk script) XCAST6 socket (C code) single C code C compiler XCAST6 whiteboard executable binary xwbd.c & xwbd.h gcc –o xwbd xwbd.c./xwbd
17 Whiteboard Data Format whiteboard command ID - sequence number timestamp
21 3. Evaluation 1.Distinguish response time of XCAST6 Whiteboard and MBONE Multicast Whiteboard 2.Routers forwarding time between XCAST6 and Multicast 3.Bandwidth consumption
22 Evaluation Process Design a network for the test. Collect byte sizes of all whiteboard commands. Determine maximum number of destinations. RTT (Round Trip Time) method to obtain response time between sender and receiver(s). Obtain routers forwarding time with Ethereal Bandwidth measurement
25 Possible number of destinations headerdestination addresses95 bytes data MTU 1500 bytes 40 bytes Up to 78 destination addresses In this remaining space
26 Response Time Response time = the time taken by a sender to send a whiteboard command to a receiver. Previously NTP (Network Time Protocol) method was used but it did not give good accuracy as response time in the testbed is in a matter of milliseconds. RTT (Round Trip Time) value divided by 2 is the best known value for the response time assuming the symmetric route path.
31 Conclusion 1.XCAST6 whiteboard is effective for a group of 20 or less in terms of response time, regardless the amount of bandwidth. An XCAST6 group of 5 or less will consume less bandwidth than Multicast. 2.Routers forwarding time between XCAST6 and Multicast differs by microseconds. No significant overhead cost difference in the router. 3.Whiteboard packets are infrequently transmitted. A more frequent transmission of XCAST6 applications such as video and audio stream may show significant difference.