Kawasaki, Japan, May 15, 2015

Fujitsu Laboratories Ltd. today announced its development of the industry's first technology that estimates the number of development work-hours needed to automate systems operations, enabling assessment of the cost effectiveness of such automation work to be generated in a short period of time.

The process of automating operations involves developing operations-task components so that individual operations can be automated, and creating workflows to control them. Ordinarily, work-hours are estimated based on descriptions of tasks in an operations manual, but when a single operations-task component appears in several descriptions or there are procedures in common, it is difficult to properly estimate the number of operations-task components requiring development and the length of the workflow, which hinders the accuracy of the estimate for development work-hours.

Fujitsu has developed a technology that accurately estimates the number of work-hours needed to develop operations-task components, by estimating the number of operations-task components based on type-identifying rules derived from approximately 1,300 patterns of task descriptions, and applying estimated development work-hours for each operations-task component. Another technology automatically extracts sections that can be split off into subroutines and estimates the length of the workflow, in order to evaluate the number of development work-hours. Combined, for systems operations involving more than 1,000 tasks that are candidates for automation, these technologies can provide a rough estimate of development work-hours in roughly five days, whereas before this process took more than a month.

This technology enables easy calculation of cost effectiveness based on the number of development work-hours and the operations task work-hours that will be obviated by automation, so that deployment plans can be devised more quickly and with overall efficacy in mind by, for example, first applying automation to tasks offering the highest cost benefit.

Details of this technology are being presented at the Workshop on Advanced IT Service Management (BDIM 2015), opening May 15 in Ottawa, Canada.

Background

Business systems are evolving into more diverse forms to adapt to specific needs, including server consolidation, redundancy between sites for disaster mitigation, and operations run on private clouds. When running operations in the cloud, for example, there are operational tasks specific to the cloud for configuring and managing resources, but the general trend in operations management is to reduce costs. To automate operational tasks that had been done by hand to achieve both a reduction in operational costs and a reduction in operator errors, and thereby improve the quality of operations, Fujitsu Laboratories in 2014 developed a technology() that analyzes operating procedures in an operations manual to present candidates for automation from among the operational tasks in it.

Issues

Taking a systematic approach to operations automation based on cost effectiveness requires an estimate of the work-hours that will be needed to develop the automation. Work-hours to develop automation can be expressed as a total of the work-hours to develop the operations-task components that will be automated for each task, and the work-hours to write the workflows that will control them.

In the process of automatically estimating work-hours based on the descriptions of tasks in an operations manual, estimating the number of operations-task components to be developed can be difficult when there is a single operations-task component mentioned in multiple descriptions of GUI actions, for parameter settings or option selections. This is a factor in amplifying errors when estimating development work-hours for components. With common procedures that could be treated as workflow subroutines, simply converting each common procedure into a subroutine will increase the subroutine calls, and may increase the total volume written. Without the ability to carefully estimate the length of the written workflow, the accuracy of the work-hours estimate will suffer.

About the Technology

Fujitsu has developed technologies that accurately estimate the work-hours needed to develop operations-task components and to generate workflows. Key features of the technologies are as follows:

1. Estimate work-hours required to develop operations-task components

By looking at the descriptions of tasks in the operations manual and consolidating task-description sections of the same type into single task components, this technology is able to accurately estimate the number of operations-task components to be developed, even when a single operations-task component is addressed by multiple descriptions, as with GUI controls. Here, depending on the details of the operations-task component, it is possible to estimate development work-hours based on previous automation work, for an accurate estimate of development work-hours (Figure 1).

To determine the type of task description, the technology checks against type-identifying rules that specify that task descriptions are the same when they contain identical or related grammatical objects. The type-identifying rules were generated based on operations manuals for 50 projects, from which some 300 types and 1,300 patterns of task descriptions were extracted. These are able to cover roughly 90% of all task descriptions, and even when a direct rule comparison is not possible, a task description can be combined with the preceding or following description to be processed as a single operations-task component.

2. Estimate work-hours required to generate workflows

When an operations manual contains common procedures, this technology calculates the savings from converting to subroutines and the overhead of adding subroutine calls, based on the length of the procedures and the number of common procedures. It calculates the work-hours in both scenarios and compares the two, and decides to use subroutines for common procedures when the total work-hours are shorter. It estimates the length of the workflow created based on the subroutine, and calculates the work-hours needed to write the workflow (Figure 2).

Results

In internal testing of operations on a large-scale system, where there were more than 1,000 candidate tasks for automation, manually estimating the work-hours needed for automation using sampling took one month or more. In contrast, these technologies were able to generate a preliminary estimate of the development work-hours needed for automation in about five days. In addition, when comparing these estimates to the actual hours worked, the average error margin was 7.0% for operational tasks that were primarily command-driven, and 19.3% for GUI-driven tasks, as compared with 65.7% using existing methods.

Calculating the period, to recover investment based on the automation development work, which is the total of the work-hours to develop operations-task components and to develop workflows, and the operational work otherwise done by hand and obviated by automation, shows that the shorter the period, the greater the cost efficiency (Figure 3). This technology makes it possible to systematically deploy automation for the greatest impact by prioritizing automations that have shorter times required to recoup their required investment.

Future Plans

Fujitsu Laboratories plans to continue working to improve the accuracy of estimates, particularly for GUI-driven tasks, with the goal of a practical implementation during fiscal 2015.

distributed by