A Replica Based Co-Scheduler (RBS) for Fault Tolerant Computational Grid - Cloud, Grid and High Performance Computing: Emerging Applications - page 112

Table 9. Final allocation of the job to the nodes

Node on which allocation has been made

m 00 P 4

m 10 P 4

m 20 P 1

m 30 P 2

m 40 P 0

Table 10. Detection of sick and healthy nodes

Node on which allocation has been made Replacement Node

m 00 P 4 (S) P 2

m 10 P 4 (S) P 0

m 20 P 1 (H) P 1

m 30 P 2 (H) P 2

m 40 P 0 (H) P 0

Table 11. New allocation of the job to the nodes

in the order P 2 , P 0 , P 1 , P 2 and P 0 for the modules
m 00 , m 10 , m 20 , m 30 and m 40 , the turnaround time
offered for various modules by the corresponding
nodes is as shown in Table 13.
The total turnaround time for the new alloca-
tion can thus be calculated as

Node on which allocation has been
made

m 00 P 4 (P 2 )

m 10 P 4 (P 0 )

m 20 P 1

m 30 P 2

Turnaround Time

m 40 P 0

= max (NEC ikn ) for 'k' on which allocation has
been made

Table 12. Modified allocation after node failure

= max (58, 12, 65.5) = 65.5

Node on which allocation has been made

m 00 P 2

As evident from Table 13, with node failures
the turnaround time has increased from 38 to 65.5
but still the program overcomes the glitches of
the node failures to finish the current execution
and the job execution is guaranteed. Later these
failed nodes can be eliminated from the CT of the
respective clusters, which is C 0 in this case and will
not be used for future allocation unless repaired.

m 10 P 0

m 20 P 1

m 30 P 2

m 40 P 0

old failure rate as λ th =0.005. Since node P 3 and
P 4 has been marked as a sick nodes these will not
be considered for future allocation. Since the new
allocation has suggested the allocation of modules