Information Technology Reference
In-Depth Information
Single-workload systems can be sized to accommodate the peak processing
needs of that one workload. By comparison, capacity planning for consolidated
systems is complicated by the need to understand the timing of each workload's
processing peaks and troughs. Insufficient planning risks experiencing simultane-
ous peaks of two workloads, with the possibility of insufficient processing capacity
to handle both peaks.
The processing needs of two workloads are either related or unrelated. One ex-
ample of related workloads comprises the application server and database server
of a two-tier or three-tier architecture. Whenever users submit transactions to
the application server, that workload then executes database queries. Unrelated
workloads tend to peak at different times if their peaks are short.
Some workloads are related, but in subtle ways. For example, two file servers
may not interact with each other, yet both may peak at 9:15 A . M . because most
workers begin to use their computers at that time.
This knowledge about the relationship between workloads can be used to maxi-
mize consolidation density. An example is shown in Figure 1.8. If a computer al-
ready has three workloads that peak in the morning, a fourth workload that peaks
in the evening may be a good choice for further consolidation and offers the best
financial savings.
Nevertheless, some concerns exist with such an approach. Namely, maximizing
virtualization density restricts workload mobility. If every computer is configured
with a set of workloads that utilize 85% of the system's capacity, migrating one
VE to another system may cause unacceptable system behavior. In other words,
there isn't a potential destination with enough room for the VE being moved. If
a computer is a likely migration target, you should ensure that there is sufficient
excess capacity for workloads that might potentially move there.
9.2.1.3 Problems Caused by Dynamic Workloads
Multiple dynamic workloads are more likely to expose the limitations of physical
computers described in the previous subsection.
Often a workload grows to consume all of a resource of its physical computer.
Its needs continue to grow until the pain is unbearable: Perhaps the impact on
sales is noticed, or a sufficiently large number of customers complain, or the plat-
form fails, causing unplanned service outages. Although some computers may
originally be purchased with spare capacity, eventually every computer with a
growing workload runs out of empty slots for CPUs, memory, and I/O controllers.
When the pain becomes unbearable, employees are assigned the task of identi-
fying a replacement system. This effort may involve middle or executive manage-
ment, have a significant impact on budgets, and last weeks to months. During
that time, the business's revenue may be severely (and negatively) affected by the
current system's shortcomings. Customers may begin using a competitor. The loss
of these customers—and the associated revenue—may be permanent.
 
Search WWH ::




Custom Search