5.4 Experimental result
For the moment, we shall concentrate on the experimental results of the collaborated computing system. In this thesis, we will take the process time and the time consumption to be the targets of analysis. The initial parameters of the experimentation are the process time less than 30% and the waiting time of 3 seconds.
Before we move to the discussion of process time and time consumption, two scenarios have to be mentioned. Scenario one is the correlation program executed in
one computer. Scenario two is the correlation program executed by the collaborated computing system. Among these two scenarios, the client computer in scenario two is the same computer as scenario one. Let us move our attention to discuss the process time.
z Process time
As we can see in Figure 5.9, the process time in scenario one is always 100%
busy, which is shown as the dotted line, but the process time in the client computer in scenario two is always under 40%. In other words, the loading in client computer has been successfully shared with other computers.
0
Figure 5. 9. The process time.
(size of subtask is 250 sets of filters, process time is 30%, waiting time is 3 seconds )
Let us take a look at the process time in servers. Figure 5.10 shows us the process time in the period while computing a sub-task. (a), (b), and (c) were the process time in server 1, 3, and 4. (d) was the process time in server 5.
All process time of servers are similar in Figure 5.10. The only difference is the period of time while finishing a sub-task. In (a),(b),(c), one sub-task can be finished within 9 minutes, which including the time of sending the sub-task, the time of computation, and the time of conveying the result. In (d), one sub-task was finished within 13 minutes.
0
(size of subtask is 250 sets of filters, process time is 30%, waiting time is 3 seconds )
The collaborated computing system aims to share the loading. Figure 5.9 and Figure 5.10 show the shared computation load.
z Time consumption
The Figure 5.11 tells us that, in the given example, the more servers involved,
the more time can be saved. Through the collaborated computing system, the time consumption is reduced dramatically while the number of servers is increased from 2 to 5. When comparing scenario one with scenario two, it is clear that the time consumption for scenario one, 47 hours and 19 minutes, is much longer than that of scenario two.
Figure 5. 11. The time consumption in the collaborated computing system.
(size of subtask is 250 sets of filters, process time is 30%, waiting time is 3 seconds )
Figure 5.11 tells us that the collaborated computing system can really share the loading. However, the speed of reduction slows down while the number of computers increases. The possible reasons are that the network is busy in transportation and the hardware configuration of servers is different. The solution is to use at least two ports for Servlets, one for sending and one for receiving, to improve the transmission efficiency. However, it is still impossible to decrease the computation time in proportion due to the hardware configuration of servers.
6 Discussion
The main objective in this thesis is to provide an inexpensive and efficient solution for load distributing. Therefore, the collaborated computing system, which is one kind of CPU power sharing P2P system, is presented. After practicing the collaborated computing system with an example in DWT based digital watermark, the analysis of this system is discussed as following.
In the collaborated computing system, the process time and the time consumption have been improved greatly by comparing to execution the analysis in single computer. Moreover, speedup is gradually ascending and efficiency is always above 0.8, while the amount of computers is up to 40. That proves the performance of the collaborated computing system is highly expected.
The collaborated computing system can solve the mentioned weaknesses and disadvantages of P2P system in Loo (2003), which are security, motivation, performance efficiency and compatibility. The collaborated computing system was proposed as a solution for sharing the loading in a small-scale company or laboratory.
Due to this proposal, the security and motivation can be easily conquered by the policy in company or laboratory. In addiction, the collaborated computing system is implemented by Servlet, which can ease the problems in security, performance efficiency, and compatibility.
This collaborated computing system can also be improved by the following approaches:
z Uses the other programming language to implement the trigger: the trigger was built by Microsoft .Net, which means that it can only run on the computer with Microsoft Operation System.
z This collaborated computing system can be implemented into the other task by modifying the job dispatch algorithm.
z Allocates jobs based on the servers’ information: the client will arrange the job loading to each server depending on their hardware configuration recorded in the client’s records.
Furthermore, there are three differences between the power server model and the proposed collaborated computing system. First, the functionalities in the coordinators in two systems are different. The coordinator in the collaborated computing system is not responsible to convey the Servlet and the correlation program between client and server. The Servlet and the correlation program are delivered by the Intranet. Hence, the coordinator can be omitted. Second, the computation application is run in Internet Explorer compared to be executed as screensaver. Due to the difference, the computation process in Internet Explorer mode can be made faster than the computation process in screensaver mode. In addition, the servers can cope with their job as usual. Third, the collaborated computing system provides interaction between the client and servers. Servers can decide how to share their process time with client.
The collaborated computing system, which is suitable for a small and medium size project, can be easily and quickly implemented by Servlet with low cost.
According to this system, computers in an enterprise or a laboratory can devote their computation power while they are idle. Therefore, by integrating those computation
powers within the organization, the time consumption for a particular project can be reduced effectively.
Reference
1. Anderson, D. P., Cobb, J., Korpela, E., Lebofsky, E. and Werthimer, D.
(2002). SETI@home: an experiment in public-resource computing.
Communications of the ACM, 45(11), 56-61.
2. Cox, I. J., Kilian, J., Leighton, F. T. and Shamoon, T. (1997). Secure spread spectrum watermarking for multimedia. IEEE Transactions on Image Processing, 6(12), 1673-1687.
3. FightAIDS@home (2004, June). Retrieved June 7, 2004 from the World Wide Web: http://fightaidsathome.scripps.edu/.
4. Foster, I., Kesselman, C. and Tuecke S. (2001). The anatomy of the Grid:
Enabling Scalable Virtual Organization. The International Journal of High Performance Computing Applications, 15(3), pp. 200-222.
5. Joseph, J. (April 7, 2003). A developer’s overview of OGSI and OGSI-based Grid computing: get an in-depth look at the Open Grid Service Infrastructure. Retrieved June 18, 2004 from the World Wide Web:
http://www-106.ibm.com/developerworks/grid/library/gr-ogsi/
6. Loo, A. W., Bloor, C. and Choi, C. (2000). Parallel computing using web servers and Servlets. Internet Research, 10(2).
7. Loo, A. W. (2003). The future of peer-to-peer computing. Communications of the ACM, 46(9), 57-61.
8. Looseley, C. and Douglas F. (1998). High performance client/server: a guide to building and managing robust distributed systems. New York: John Wiley & Sons.
9. Lui, S. M. and Kwok, S. H. (2002). Interoperability of peer-to-peer file sharing protocols. ACM SIGecom Exchanges, 3(3), 25-33.
10. Napster Inc. (2004, June). Retrieved June 7, 2004 from the World Wide Web: http://www.napster.com.
11. Talia, D. and Trunfio, P. (2003, July/August). Toward a synergy between P2P and Grids. IEEE Internet Computing, 93-95.
12. Tseng, L. H. (2003). The wavelet packet transform based watermarking for the digital image ownership verification. Unpublished master’s thesis, National Chiao-Tung University, Taiwan.
13. Wang, Y., Doherty, J. F. and Dyck, R. E. (2002). A wavelet-based watermarking algorithm for ownership verification of digital images. IEEE Transactions on Image Processing, 11(2), 77-87.
14. Zhang, L. J., Chung, J. Y. and Zhou, Q. (October 1, 2002). Developing Grid computing applications, part 1: introduction of a Grid architecture and toolkit for building Grid solutions. Retrieved June 18, 2004 from the World Wide Web:
http://www-1.ibm.com/servers/eserver/central/strategic_direction.html?ca=e servercentral&me=W&met=9_to_sa_strategicd