Hardware Management - Deploying and Managing a Cloud Infrastructure - page 232

that a storage cluster's capacity is 1 petabyte or that a system has 128 teraflops of computing
power. This simply makes it easier to distribute and collect the resources if we view it as a
single pool, and it's also easier to manage.
The idea here is that resources can be dynamically allocated from these large resource
pools while idle resources previously allocated to tenants can be reclaimed and returned to
the pool. Tenants can dynamically scale with the workloads they have as long as the pool
has contents. The management of the resources would be abstracted from the tenants, who
would simply perceive the pool to be limitless.
Proper Virtual Resource Allocation (Tenant/Client)
For the host, resource allocation is all about proper management, allocation, and recollection
to keep the resource pools filled. At the other side of the equation is the tenant. Abstracted
from what the underlying hypervisor is doing, the tenant view on resource allocation is about
scalability to keep up with workloads while keeping things lean in order to keep costs down.
Virtual CPU
The vCPU is often the most controlled among the resources because it is the most finite
and most expensive. Service providers often impose hard limits for the deployment for
each VM instance since this is the only resource that cannot be literally divided among
users. As previously discussed, each VM instance is given only a certain amount of CPU
time based on the vCPU allocated for it.
The amount of vCPU that can be assigned to a VM usually depends on the service tier
or package that you have licensed. For example, most service providers would probably
allow a single dual-core vCPU for their cheapest standard package and more than eight
vCPUs for their most-expensive package. So it goes without saying that you should choose
the service package most suitable to your workflows; do not go lower or higher. Lower-cost
packages would not offer enough performance and capacity, while higher-cost ones might
not offer enough benefits because some of the resources are underutilized.
Memory
As with every virtual resource, virtual memory limits will be imposed for every service
package. The cheaper the package, the less available memory provided. The usual scheme
for memory allocation is like that for vCPU, where you have to specify the amount of
memory for each instance.
In recent years, some service providers have begun offering virtual memory as resource
pools that tenants can make use of in any way they want. This allows for great flexibility
because you can allocate more memory to specific instances and less to others, so you can
produce very lean virtual machines.
Storage and Network Allocation
Just like physical computers, virtual machines are created for different purposes. Some
are created for computation, others are for storage (such as database servers), and others are