Kun-Ting Chen, Chien Chen, Po-Hsian Wang
Presented by: Katya Rodriguez, Ahmed
Alsuwat, and Saud Tawi
3.11.15

Introduction

Related Works and Motivations

Problem

Network-Aware Bipartite Matching Load
Algorithm

Contribution to the Field

Questions?

Cloud data centers use virtualization-based
technology for the sole purpose of
consolidating hardware resource usage.
◦ This provides application hosting for multiple
service providers.

Without proper allocation, the loads of
different resources may become unbalanced
among different physical hosts.


Currently, the existing algorithms for load
balancing search for a VM to begin a
migration.
The selection of the next migration doesn’t
occur until after the previous migration has
completely ended.
◦ Also known as sequential migration

It proposes a different method of migrating
data centers in a VM.

This method doesn’t degrade application
performance.

Speed up VM migration for data centers by
using a network-topology that is aware of
parallel migration.
◦ Network-Aware Bipartite Matching Load-Balancing
Algorithm



Many of the previously proposed load
balancing schemes for cloud computing
measure the load on physical hosts
differently.
Zhao and Huang use the number of VMs of a
host as their load measurement.
Mimicking animal behavior of honeybees for
load balancing.

Considering the demands of the VM and
heterogeneous capacity in each host.
◦ Have overloaded hosts below their threshold by
migrating VMs sequentially.

Even though currently the modern cloud
administrators migrate multiple VMs
concurrently, migration still has a chance to
degrade into sequential migration.
Dimension of resources consumed by a VM = vmb
VM host = ha
Trigger set = TR
Vector of system threshold along each dimension of resources = t
Vectors of capacity and resource utilization along n dimensions of
resources = u
Standard deviation of resource usage of the hosts is least in the case of Network-Aware
Bipartite matching (NABM). Standard deviation of VectorDot (VD) and NOLB is greater than
the standard deviation of (NABM).
‣ Network-Aware Bipartite matching
(NABM) allows the system to strike
a balanced state quicker than
allowed by VectorDot (VD).
‣ The graph states that as the
number of hosts is increased, the
time taken by NABM to reach a
balanced state remains more or less
the same, but the time taken by VD
increased almost proportionally.
‣ VM migrations per round is one
in the case of VD, but in the
case of NABM, average VM
migrations increase in
proportion to the addition of the
number of hosts.
‣ This result unveils that
performance of parallel VM
migration load management
system improves as more hosts
are added to the system, but it
does not happen in the case of
sequential load management
system.
‣ This result implies that parallel
VM migration system handles
overloaded hosts more
effectively than sequential
system.
‣ This is because the algorithm of
NABM does not break down
when hosts are overburdened.
Instead, it removes the queues
quickly and allows the users to
get information from servers
without delays.
‣ If network parameter value
goes beyond .6, then
performance level may not
remain the same even under
parallel VM migration.


Viability and usefulness of the parallel VM
migration for data centers in managing load.
Existing load management systems of virtual
data centers do not ensure reliable
throughput, swift handing of job queues by
servers and fast attainment of balanced state.

Handles multiple migrations of VMs at one
time
◦ Reducing downtime
◦ Increasing service efficiency

By adapting parallel VM migration, the data
can be migrated from one center to another
very quickly.
◦ It enhances the ability of the cloud computing
technology to foil data theft attempts.

A sustained balanced state, under parallel VM
migration, cuts down the cost of system
maintenance arising out of frequent system
breakdowns, hotspots and over usage of
resources.