While computers have become more powerful over the years, the applications we expect them to run have grown more complex even faster. As a result, it is often desirable to spread a computational workload out across a number of machines.
Further, the computational requirements of a given application can fluctuate dramatically, sometimes even from minute to minute. For this reason, many organizations find it inefficient to maintain their own hardware and prefer to rent “cloud” resources.
We will take a quick look at two related practices that facilitate the modern computing landscape: cluster computing, and cloud computing.
A cluster is, put simply, a collection of computers that operate cooperatively in some way or another. Clusters are common in scientific applications where they are used to solve very large mathematical problems that would simply take too long to complete on an ordinary workstation. But they are also used in industry to run database systems, game servers, and web applications.
For applications that must keep track of some state, such as direct messages between users, the use of clusters has motivated a great deal of work in a field called “distributed systems”. For example, if a messaging application server is run on two computers at once (two “nodes”), then it is important that a new message be recorded on both of them, regardless of which one initially received the message from the user’s device. It is also important that the message be recorded only once (you may have noticed unintended duplicates in certain apps).
As it turns out, this is a ferociously difficult problem to solve. Theoretically, there are a number of good, though sub-optimal, algorithms such as Paxos, but “distributed consensus”, as this problem is called, remains difficult in practice.
We will take a tour of one particular application of a high performance computing cluster and see a cluster in action, including how tasks are configured and how output is collected upon completion.
Cloud computing, in this context, refers mostly to Infrastructure as a Service (IaaS) platforms such as Amazon Web Services (AWS). IaaS providers allow organizations to “rent” virtual hardware for various purposes. Many systems even allow the creation of entire virtual data centers complete with custom network topology and firewalls.
The primary advantage IaaS vendors offer is flexibility. It is generally possible to “spin up” a new machine in a minute or less. This means that as requirements change, so can computing capability. It is also possible to assemble an ad hoc cluster this way, which is particularly nice for organizations that have only infrequent need for massive computing power.
We will create a couple cloud servers in class and look at some of the ways they can be configured. We will also explore some of the other offerings available through AWS.