**7. Experiments**

In order to test this model, we built a *DNE* that is composed of 2 file servers and 8 cache nodes (personal computer), and 16 client computers, and they are connected by LAN. Each server has 4 disks, and the segments of large file are distributed in all disks in balance. There are 4 files and the length of all them is about 1GB. The segment length is 48MB. The cache nodes are classified into 2 types according to their types of cpu, memory, disk, and net adapter. The types of computers are *RSV* (2400MHZ, 768MB, 7200RPM, 100M) and *RSV* (1800MHZ, 512MB, 7200RPM, 100M). The operating systems of the computers are the LINUX. In order to test the peak access ability about the hot segment, we ignore the net factors. So, the test agent running on client end only sends the access command timely and the *CA* does the file operations and the data will be not transferred to the client end. The average value of *d*-density allocation strategy is 6.

**Experiment 1.** The experimentation includes two kinds: the one is the response time testing in the condition that all *DSEG*s are stored in the file server disks; the other is the response time testing in the condition that all *DSEG*s are stored in Grid memory. We calculate the average ratios of response time of this two kinds. The tests include 6 times according to the *DSEG* scales. The total number of *DSEGs* is the *Tc*, 2*Tc*… 6*Tc* during the 6 times testing. In each test, we adopt the three access method for the *DSEG*s: the random access, the sequence access, and the mix access composed of the random and sequence. The test results are shown in figure 6(a), and the results show that PCMGM is high efficiency compared with the disk cache method.

**Experiment 2.** We had tested the variety of PCMGM performance during d is 1, 2, 3, 4, 5, and 6 (The total number of DSEGs is the Tc, 2Tc… 6Tc). The test results are shown in figure

Research and Implementation of Parallel Cache Model Through Grid Memory 149

We would like to thank the support of National Nature Science Foundation of China (No.60573108), Shanghai Leading Academic Discipline Project (No.T0502) and Shanghai

Anurag Acharya and Sanjeev Setia: Using Idle Memory for Data-Intensive Computations. In

Craig S. Freedman, Josef Burger, and David J. Dewitt: SPIFFI-a Scable Parallel File System

E. Osawa: A Scheme for Agent Collaboration in Open MultiAgent Environment .Proceeding

E. P. Markatos and G. Dramitions: Implementation of Reliable Remote Memory Pager. In proceedings of the 1996 Usenix technical Conference, (1996):177-190 Horst Eidenberger: Gigabit Ethernet-Based Parallel Video Processing, Proceedings of 11th

I. Foster, C. Kesselman: The Grid: Blueprint for Future Computing Infrastructure. San

J. Fernandez,J. Carretero, F. Garcia-Carballeira, A. Calderon, J. M. Perez-Menor: New Stream

J. O. Kephart, Chess, D. M: The Vision of Autonomic Computing, IEEE Computer, (2003):

J. Carretero, F. Perez, P. de Miguel, and L.Alonso: ParFiSys: A Parallel File System for MPP.

O. F. Rana and D.W. Walker: The Agent Grid: Agent-based resource integration in PSEs, In

P. F. Corbett and D. G. Feitelson: The Vesta Parallel File System. ACM transaction on

Pal Halvorsen, Carsten Griwodz, Vera Goebel, Ketil Lund, Thomas Plagemann, Jonathan

Rajkumar Buyya: High performance Cluster Computing Architectures and Systems,

Seogyun Kim, Jiseung Nam, Soon-ja Yeom:Effective delivery of Virtual Class on Parallel

Distribution (AXMEDIS'05),IEEE Computer Society,(2005): 77-84

mathematics and Simulation, Lausanne, Switzerland, (2000 )

proceedings of the 1998 ACM SIGMETRICS Conference on Measurement and

for the Intel Paragon. IEEE Transactions on Parallel and Distributed System, Vol.7

International Multimedia Modelling Conference (MMM'05), IEEE Computer

Caching Schemas for Multimedia Systems, In Proceedings of First International Conference on Automated Production of Cross Media Content for Multi-Channel

proceedings of the 16th IMACS World congress on Scientific Computing, Applied

walpole: Storage System Support for Continuous Media Applications, Part2: Multiple Disks, Memory, and Integration, IEEE Distributed systems online 1541-

Media Stream Server, Proceedings of the International Conference on Computers in

Education Committee Foundation (No. 06QZ002, 07ZZ92).

Modeling of Computer Systems, (1998):278-279

Francisco, USA: Morgan Kaufmann Publishers, (1999)

Operating System review, Vol.30 (2), (1996):74-80

Computer Systems, Vol.14 (3), (1996):225-264

**9. Acknowledgement** 

(11), (1996):1185-1200

Society, (2005):358-363

4922,vol.5(2), (2004)

Prentice Hall, (1999)

Education, IEEE press,2002

41-50

of IJCAI'93, (1993):352-358

**10. References** 

6(b), and the results show that the key influence factor about PCMGM performance is the capacity of Grid memory. So, mining large scale idle computer to construct the grid memory can improve the concurrent access response time for the hot segments during the peak access.

Fig. 6. The test results
