Academia.eduAcademia.edu

Figure 1 – uploaded by Paul Millar

See full PDFdownloadDownload figure
Table 2. Estimated sizes of CDF secondary data sets (from [12]). There will be some distribution of jobs each site performs. In the simulation, we modelled this distribution such that each site ran an equal number of jobs of each type except for a preferred job type, which ran twice as often. This job type was chosen for each site based on storage considerations; for the replication algorithms to be effective, the local storage on each site had to be able to hold all the files for the preferred job type.

Table 2 Estimated sizes of CDF secondary data sets (from [12]). There will be some distribution of jobs each site performs. In the simulation, we modelled this distribution such that each site ran an equal number of jobs of each type except for a preferred job type, which ran twice as often. This job type was chosen for each site based on storage considerations; for the replication algorithms to be effective, the local storage on each site had to be able to hold all the files for the preferred job type.

Related Figures (5)

Fig. 1. Simulated DataGrid Architecture. One of the main design considerations for OptorSim is to model the interactions of the individual Grid components of a running Data Grid as realistically as pos- sible. Therefore, the simulation is based on the architecture of the EU DataGrid project [14] as illustrated in Figure 1.
Fig. 1. Simulated DataGrid Architecture. One of the main design considerations for OptorSim is to model the interactions of the individual Grid components of a running Data Grid as realistically as pos- sible. Therefore, the simulation is based on the architecture of the EU DataGrid project [14] as illustrated in Figure 1.
Fig. 2. The EU DataGrid TestBed 1 sites and the approximate network geometry. The numbers indicate the bandwidth between two sites. in Figure 2. Within this model, each site was allocated storage resources propor- tional to their actual hardware allocations. Each TestBed site, excluding CERN, was assigned a Computing and Storage Element. CERN was allocated a Stor- age Element to hold all of the master files but was not assigned a Computing Element. Routers, as previously stated, were described by creating a site with- out Computing or Storage Elements. The size of the Storage Elements for each TestBed site are given in Table 1.
Fig. 2. The EU DataGrid TestBed 1 sites and the approximate network geometry. The numbers indicate the bandwidth between two sites. in Figure 2. Within this model, each site was allocated storage resources propor- tional to their actual hardware allocations. Each TestBed site, excluding CERN, was assigned a Computing and Storage Element. CERN was allocated a Stor- age Element to hold all of the master files but was not assigned a Computing Element. Routers, as previously stated, were described by creating a site with- out Computing or Storage Elements. The size of the Storage Elements for each TestBed site are given in Table 1.
Table 1. A list of resources allocated to the TestBed 1 sites, from which the results it this paper were generated.
Table 1. A list of resources allocated to the TestBed 1 sites, from which the results it this paper were generated.
Fig. 3. A histogram of job duration (left) and the progression of job duration over course of the simulation (right).
Fig. 3. A histogram of job duration (left) and the progression of job duration over course of the simulation (right).
Fig. 4. Integrated running times for 10000 jobs using each access pattern and replice optimisation algorithm. The variation of job duration over the simulation is shown in the right his- togram in Figure 3 for the same job type and Computing Element as above. There is clearly a large variation in the job duration due to the factors already mentioned, but the general trend is for jobs to be executed more quickly over time, indicating the movement toward a more optimal replica configuration.
Fig. 4. Integrated running times for 10000 jobs using each access pattern and replice optimisation algorithm. The variation of job duration over the simulation is shown in the right his- togram in Figure 3 for the same job type and Computing Element as above. There is clearly a large variation in the job duration due to the factors already mentioned, but the general trend is for jobs to be executed more quickly over time, indicating the movement toward a more optimal replica configuration.
Related topics:
Distributed ComputingGrid ComputingData ReplicationComputational GridData Grid
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved