Information Technology Reference
In-Depth Information
Figure 6. SALB algorithm
Experimental Environment
around time is chosen to be the destination site r .
Finally, T jt for each site is compared. If T jt of site
r is less than that of the site l , site l will migrate
job J to site r , otherwise site l retains job J until
the next time interval.
In this experiment, we adopts JXTA with version
2.5.1, Java with version 1.6.0 and Condor with
version 6.7.20 to implement SALB in Taiwan
UniGrid for evaluating load balancing. In addition,
we construct five grid sites, and each site consists
of one super node and some general nodes. Super
nodes are responsible for the communication with
other neighbor sites, assigning jobs to their general
nodes and executing SALB strategy. General nodes
are responsible for executing the jobs assigned
by the super nodes. Moreover, SALB can be
extended to larger scale systems. Table 3 shows
the specification of the experimental platform.
EXPERIMENTAL RESULTS
This section introduces the experimental environ-
ment and results of SALB. In our discussion of
the experimental results, we particularly focus on
the efficiency of load balancing.
Table 3. System specification
Site
Hosts
Peer Types
CPU clock
Memory
1
Host201
Super node
Intel P-D 3.40GHz x 2
512M
1
Host204
General node
Intel P-D 3.40GHz x 2
512M
2
Host205
Super node
Intel P-4 3.40GHz x 2
512M
2
Host208
General node
Intel P-4 3.40GHz x 2
512M
3
Host206
Super node
Intel P-4 3.40GHz x 2
512M
3
Host207
General node
Intel P-4 3.40GHz x 2
512M
4
Host221
Super node
Intel P-4 3.40GHz
256M
4
Host223
General node
Intel P-4 3.40GHz
256M
5
Host222
Super node
Intel P-4 3.40GHz x 2
512M
5
Host224
General node
Intel P-4 3.40GHz x 2
512M
 
Search WWH ::




Custom Search