Selecting a Service Platform - The Practice of Cloud System Administration - page 49

Terms to Know
Server: Software that provides a function or API. (Not a piece of hardware.)
Service: A user-visible system or product composed of many servers.
Machine: A virtual or physical machine.
Oversubscribed: A system that provides capacity X is used in a place where Y
capacity is needed, when X < Y . Used to describe a potential or actual need.
Undersubscribed: The opposite of oversubscribed.
Although the service provider manages its layers of the infrastructure, IaaS does not re-
lieveyoufromallwork.Alotofworkmustbedonetocoordinateallthepieces,understand
and tune the various parts so they work well together, and manage the operating system
(since you have total control of the OS).
Providers charge for compute time, storage, and network traffic. These costs will affect
how your application is architected. Keeping information locally versus retrieving it over
a network may have different costs, affecting your design choices. If information is ac-
cessedfrequentlyoverthenetwork,thenetworkchargescanbereducedbycachingorstor-
ing more information locally. However, the additional local storage may have its own cost.
Such engineering details are important, because otherwise you may find yourself with a
startlingly large bill at the end of the month. These are important points to work out with
the development and business teams. The software and operational choices have real costs
and tradeoffs.
The performance characteristics of providers may vary wildly. When comparing pro-
viders, it is important to benchmark local storage, remote storage, CPU, and network per-
formance. Some providers' remote storage is significantly faster than others. Repeat any
such benchmarks at different times of the day—some service providers may experience
high packet loss at daily peak times. Design decisions made for one provider may not be
the right choice for other providers.
Within an IaaS offering, partitions or “reliability zones” segment the service geograph-
icallytoprovideregionaluptimeguarantees.Whileallattemptsaremadetoensurereliable
service, it is inevitable that some downtime will be required for maintenance or due to un-
avoidable circumstances suchasanatural disaster.Theservice providershouldsegment its
serviceintomultiplezonesandprovideguaranteesthatplanneddowntimewillnotoccurin
multiple zones at the same time. Each zone should be far enough apart from the others that
naturaldisastersareunlikelytostrikemorethanonezoneatatime.Thispermitscustomers
to keep their service in one zone and fail over to another zone if necessary.