Hardware Management - Deploying and Managing a Cloud Infrastructure - page 230

more-powerful virtual machines, they have to avail of more-expensive offerings or pay for
extra resources used.
Network bandwidth allocation is a different matter because unlike CPU, memory, and
storage, it is not proportionally shared according to payment, nor does any provider offer
minimum guarantees on network bandwidth. This resource is difficult to share because the
network allocation for internal virtual machines depends on its connections from outside
networks, which are out of the provider's control. And as part of the quality of service agree-
ment, the service provider has to offer fast and satisfactory access to the virtual machines and
the cloud, which means full access of the network. But in the cloud computing paradigm we
should be able to proportionately share this resource.
First you need to fulfill some requirements.
Guaranteeing Minimum Bandwidth One is that you should be able to provide each ten-
ant or client with a minimum guarantee for network bandwidth that they can expect for
each VM they have without regard for the other tenants' network utilization. This is the
minimum guarantee.
Providing High Utilization Performance Another requirement to be met is to be able
to provide high utilization performance, which is providing more network bandwidth to
applications with high utilization in terms of bandwidth when other tenants are actually
using less. This will significantly improve the performance of applications that experience
a dynamic traffic flow, which may increase or decrease dramatically rather quickly.
There are two ways to tackle this problem, but there are trade-offs. If you follow a mini-
mum guarantee path, you cannot provide proportionality because every VM, client, or ten-
ant will be allocated bandwidth whether they are using it or not. This leads some bandwidth
being underutilized. However, if you implement proportionality in allocation, you may not
be able to guarantee minimum network performance at all times because you will need to
use a first-come, first-served model. Tenants who have been network-idle and have not been
allocated minimum bandwidth will experience little or no connectivity, at least for a short
time until proportional resource allocation has finally run its course. There are downsides
to everything, so you just have to find the proper balance. A hybrid approach might work,
making the client decide whether to opt for proportional or minimum guarantee depending
on their needs, which will then be reflected in the service-level agreement (SLA).
Entitlement/Quotas (Shares)
For a host or service provider, the most obvious resource management measure related to
tenants is the imposition of quotas, which should be a clearly defined line stating how large
a share each one is entitled to. This makes it easy to actually determine the actual tenant
capacity of the infrastructure in terms of license or service tiers. So even without the use of
advanced resource sharing algorithms in the hypervisor, you get a clear picture of the full
extent of the number of tenants that can be supported satisfactorily.
Let us go back to our comparison of a multitenant cloud environment and an apartment
building. Just like an apartment building, a cloud service will have entitlements that can be
likened to rooms, which differ in size and amenities, so the owner (service provider) knows