CSci 555: Functional Programming
Spring 2016

Effect of Computing Hardware Evolution on Programming Languages

Note: These brief notes are expanded from the handwritten notes used by the instructor for the lecture/discussion on this topic in CSci 450/503 (Organization of Programming Languages) on 27 August 2014.

When were the first "modern" computers developed? That is, programmable electronic computers.

Although the mathematical roots of computing go back more than a thousand years, it is only with the invention of the programmable electronic digital computer during the World War II era of the 1930s and 1940s that modern computing began to take shape.

One of the first computers was the ENIAC (Electronic Numerical Integrator and Computer), developed in the mid-1940s at the University of Pennsylvania. When construction was completed in 1946, the ENIAC cost about $500,000. In today's terms, that is more than $5,000,000. It weighed 30 tons, occupied as much space as a small house, and consumed 160 kilowatts of electric power.

The ENIAC and most other computers of that era were designed for military purposes, such as calculating firing tables for artillery or breaking codes. As a result, many observers viewed the market for such devices to be quite small. The observers were wrong!

Electronics technology has improved greatly in 70 years. Today, a computer with the capacity of the ENIAC would be smaller than a coin from our pockets, would consume little power, and cost just a few dollars on the mass market.

How have computer systems and their use evolved over the past 70 years?

• Contemporary processors are much smaller and faster. They use much less power, cost much less money, and operate much more reliably.
• Contemporary "main" memories are much larger in capacity, smaller in physical size, and faster in access speed. They also use much less power, cost much less money, and operate much more reliably.
• The number of processors per machine has increased from one to many. First, channels and other co-processors were added, then multiple CPUs. Today, computer chips for common desktop and mobile applications have several processors--cores--on each chip, plus specialized processors such as graphics processing units (GPUs) for data manipulation and parallel computation. This trend toward multiprocessors will likely continue given that physics dictates limits on how small and fast we can make computer processors; to continue to increase in power means increasing parallelism.
• Contemporary external storage devices are much larger in capacity, smaller in size, faster in access time, and less expensive to construct.
• The number of computers available per user has increased from much less than one to many more than one.
• Early systems were often locked into rooms, with few or no direct connections to the external world and just a few kinds of input/output devices. Contemporary systems may be on the user's desktop or in the user's backpack, be connected to the internet, and have many kinds of input/output devises.
• The range of applications has increased from a few specialized applications (e.g., code-breaking, artillery firing tables) to almost all human activities.
• The cost of the human staff to program, operate, and support computer systems has probably increased somewhat (in constant dollars).

How have these changes affected programming practice?

• In the early days of computing, computers were very expensive and the cost of the human workers to use them relatively less. Today, the opposite holds. So we need to maximize human productivity.
• In the early days of computing, the slow processor speeds and small memory sizes meant that programmers had to control these precious resources to be able to carry out most routine computations. Although we still need to use efficient algorithms and data structures and use good coding practices, programmers can now bring large amounts of computing capacity to bear on most problems. We can use more computing resources to improve productivity to program development and maintenance. The size of the problems we can solve computationally has increased.
• In the early days of computing, multiple applications and users usually had to share one computer. Today, we can often apply many processors for each user and application if needed. Increasingly, applications must be able to use multiple processors effectively.
• Security on early systems meant keeping the computers in locked rooms and restricting physical access to those rooms. In contemporary networked systems with diverse applications, security has become a much more difficult issue with many aspects.
• etc.

The first higher-level programming languages began to appear in the 1950s. IBM released the first compiler for a programming language in 1957--for the scientific programming language Fortran. Although Fortran has evolved considerably during the past 60 years, it is still in use today.

How have the above changes affected programming language design and implementation over the past 60 years?

• Contemporary programming languages often use automatic memory allocation and deallocation (e.g., garbage collection) to manage a program's memory. Although programs in these languages may use more memory and processor cycles than hand-optimized programs, they can increase programmer productivity and the security and reliability of the programs. Think Java, C#, and Python versus C and C++.
• Contemporary programming languages are often implemented using an interpreter instead of a compiler that translates the program to the processor's machine code--or be implemented a compiler to a virtual machine instruction set (which is itself interpreted on the host processor). Again they use more processor and memory resources to increase programmer productivity and the security and reliability of the programs. Think Java, C#, and Python versus C and C++.
• Contemporary programming languages should make the capabilities of contemporary multicore systems conveniently and safely available to programs and applications. To fail to do so limits the performance and scalability of the application. Think Erlang, Scala, and Clojure versus C, C++, and Java.
• Contemporary programming languages increasingly incorporate declarative features (higher order functions, recursion, immutable data structures, generators, etc.). These features offer the potential of increasing programming productivity, increasing the security and reliability of programs, and more conveniently and safely providing access to multicore processor capabilities. Think Scala, Clojure, and Java 8 and beyond versus C, C++, and older Java.

• etc.

As we study programming and programming languages in this course--in particular functional and logic languages--we need to keep the above nature of the contemporary programming scene in mind.

UP to CSci 555 Lecture Notes root document?