Professional Documents
Culture Documents
Different Policies For Dynamic Load Balancing IJERTV1IS10485 PDF
Different Policies For Dynamic Load Balancing IJERTV1IS10485 PDF
Different Policies For Dynamic Load Balancing IJERTV1IS10485 PDF
ISSN: 2278-0181
Vol. 1 Issue 10, December- 2012
Many issues involve in dynamic load Measuring the load in distributed system
balancing such as how to measure the is very difficult task. Usually load is
load of processing elements, how should measure by metric these can be divided
load information we should collect and into two parts, Simple load indices and
where the load is reside. How the real complex load indices
component .we consider the dynamic on the load on the CPU. Simple load
load balancing algorithms consist of four indices include processor queue length
and information exchange rule, load memory available, the context switch
balancing operation, an initiation rule rate, the system call rate, CPU
distribution rule) these issues are also Complex load indices: these consist a
Key words: Load measurement, load comprise the CPU load and the amount
Load measurement policies: measuring paging rate, and the amount of ideal time
of load of every node is very important at processing node. The load indices
www.ijert.org 1
International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
Vol. 1 Issue 10, December- 2012
www.ijert.org 2
International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
Vol. 1 Issue 10, December- 2012
Where the load should be maintained? system is not vulnerable to the failure of
Central approach: central approach any single node. There are also
can be used to calculate load-state disadvantages there is no consistent
information. This is collected from all of system wide view of state, and each
the nodes in the system and made node holds different information
available when a load-sharing decision depending on which other nodes it has
must be made. Some centralized nodes communicated with, how recently that
are simply responsible for the collection communication took place, and the delay
and distributing of information, while experienced in that communication. This
others additionally act as matchmakers can lead to instability if there are
between sender and receiver nodes. significant differences in the views held.
Centralized node can work well in small How the load is collected?
or moderately sized systems, but can First option collects the load from all
become communication bottlenecks systems. Another opinion is to divide the
RRTT
when the system is scaled up. Where whole distributed systems into number
centralized components are used in the of cluster and then collect the load from
IIJJEE
entire system, that system is vulnerable different cluster then finally calculates
to the failure of the single component the over all load from all the clusters.
unless some form of backup or The choices in between these two
replication is provided, this increases extremes use local load information
complexity. collected from a certain domain of
Distributed approaches: Distributed processing nodes in which size may be
approaches are more difficult to build either fixed or variable. The global
than their centralized counterparts. The knowledge of the system’s attributes
semantics involved can be complex. (like the total work load) is prohibitive,
Each node collects information due to the communication overhead
concerning the load state at other nodes produced. This is true for large-scale
in the system. Nodes autonomously base distributed systems. Thus, the technique
load sharing decisions on the of demanding global information is
information they hold. One advantage of rejected, and partial information is used
distributed implementations is that the
www.ijert.org 3
International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
Vol. 1 Issue 10, December- 2012
www.ijert.org 4
International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
Vol. 1 Issue 10, December- 2012
than one and L is less than one. These and have no tasks available for no
two values determine the flexibility and preemptive transfer if it is polled by a
the effectiveness of a load-balancing receiver. A selection policy should
mechanism. The heavy threshold, T1, is consider at least three factors. The
computed as the product of H and L overhead incurred in transferring the
average. Similarly, the light threshold T2 task should be minimized. No
is computed as the product of L and L preemptive transfers and small tasks
average.. The transfer policy may be (small amounts of information) carry
either periodic or event-triggered. The less overhead. The execution time of the
algorithm may periodically check transferred task should be sufficient to
whether the node’s state qualifies itself justify the cost of the transfer. Even if
as a candidate for a task transfer. task execution is unknown, it should be
Selection policy: The role of selection possible to classify the tasks as short or
policy is to select tasks for transfer. In long tasks, and to consider only the long
RRTT
sender-initiated schemes, busy nodes tasks for migration. Some classification
choose tasks to transfer to another node, errors might be tolerated as load-
IIJJEE
www.ijert.org 5
International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
Vol. 1 Issue 10, December- 2012
www.ijert.org 6
International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
Vol. 1 Issue 10, December- 2012
www.ijert.org 7
International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
Vol. 1 Issue 10, December- 2012
www.ijert.org 8