2013-03-05

Remote Contractors - Lessons Learned



1. Executive Summary

A customer contracted us to update one of their applications.  Since the project was not technically challenging and involved a significant amount of repetitive effort, outsourcing was deemed to be low risk.
The project trials outsourcing development effort to remote contractors in an effort to better understand what is involved and when outsourcing is beneficial.  To reduce development costs and maximize profits, as well as the highly repetitive nature of the project, it was decided that some of the development effort would be outsourced.
The following describes project details and focuses primarily on lessons learned while working with remote contractors.

2. Introduction

2.1.    Objectives

The purpose of the project was to upgrade an existing application.  The application utilizes a couple of different technologies and programming languages.  The required updates were clearly defined by the client.

2.2.    Approach

The approach was to outsource part of the project to contract developers based in India.  This was done as a trial, with the desire to exploit outsourcing in future projects.  The project was selected as an ideal candidate for outsourcing because the effort and tasks were well known and low risk.  
It was decided to outsource as much of the development as possible, while keeping any complicated development effort internal.  This was done because the design pattern for a significant amount of the effort is quite straight-forward, so it can be easily documented, taught to contractors, and reviewed for adherence.
Other development effort was kept internal because there was a desire to increase the knowledgebase of developers within the company, adding to their familiarity with the development environment and applications.  For the project, it was important to keep the knowledge base and experience in-house, so specific less repetitive tasks were not outsourced.  We also wanted to keep the more interesting and challenging work internal to benefit employees and to ensure it was done properly.

3.     Lessons Learned

There are several lessons learned from the project, including ones specific to outsourcing effort.

3.1.    Development Environment

Due to the limited overlap of business hours, it’s very easy to waste time getting development environments working.  Originally, the plan was to host a couple of Virtual Machines (VM) locally and get the contractors to remotely access them.  This would allow us to tightly control and monitor the development environment used by contractors, as well has limiting their access to the end customer’s source code.  However, this did not work at all.  The latency that the contractors were observing made development extremely difficult.  Typing and mouse movements were likely delayed by approximately one-third of a second (ping www.pluggd.in).
To eliminate the latency issue, VMs were packaged and made available to the developers.  In my opinion, remote contractors have to use local development environments for a majority of their work to be productive. 
I highly recommend using VMs to encapsulate the development environment.  If this approach is taken, the following steps can help the process go smoothly:
  • Have the VM packaged and available well before the start date of the contractors. 
  • Transferring it will take a significant amount of time.
  • If possible, get them to transfer it over the weekend to minimize wasted hours
  • Contractors should use a user with the minimum amount of permissions required
  • Thoroughly test the VM remotely to ensure it will work for the contractors right away, and for the entire project.  It is difficult to make changes to the VM after the fact
If a development environment is going to be provided to contractors that they can install, provide a detailed step-by-step installation guide, making it as simple as possible.

3.2.    Productivity and Task Assignment

Initially, assume that contractors require very specific tasks that include detailed descriptions.  Concisely specifying tasks minimize assumptions and will ensure that they are completed as desired.  Assumptions likely differ due to differences in education, culture and project familiarity.  After a while, once becoming more comfortable with the work and capabilities of the contractors, the level of detail required in descriptions can probably be reduced.
Assume that contractors will take 3-5 days to get familiar with new development patterns.  During this period, progress will be slow.  Once familiar, the contractors will reach their full potential and productivity will peak.  This is most relevant for highly repetitive tasks.
From experience with this project, it’s safe to assume that contractors will work slower and will be less efficient than local resources.  Without doing significant investigation into the numbers, it’s assumed that the contractors were approximately 75% as productive.  This could be due to various reasons such as: language barrier; no direct supervision outside of their company; individual capabilities.  An exact number could be calculated by analyzing the estimated and actual hours of work items for various resources tracked in Rational Team Concert (RTC).
Expect to spend 30-60 minutes per day reviewing the work of the contractors to ensure quality.  Initially, 2-3 hours of review was required because there was a lot of feedback required.  This review time decreased to 30-60 minutes once they became familiar with the design pattern.
RTC was used to manage work items and as a code repository.  Although RTC has tools and workflows that are designed for reviewing and accepting developer’s code into the repository, only the reviewing functionality was used.  The reviewing functionality allowed us to efficiently examine proposed changes, including source code diffs and syntax highlighting.  The ability to accept and reject submitted source code changes would have been useful.  However, it was not used due to lack of familiarity with RTC.
Due to the learning curve, keeping contractors working on similar and repetitive tasks was most efficient.  Each time a work item with a different work flow was started there was a learning curve that had to be overcome.  This included the contractor being less productive, as well as requiring more supervision, review and feedback.  Expect the learning curve to be larger than with local resources, likely due to less overlap in business hours, indirect contact and a language barrier causing miscommunications.
Be wary whenever a new task or workflow is being performed by contractors.  Review the results closely.  The contractors tended to be good at repeating tasks, however struggled with new tasks that required undefined design patterns.  For example, they struggled with extending development patterns to other source code languages.  Design patterns had to be clearly defined and documented.  Then, the contractors were trained and their initial work closely monitored.
Getting the contractors to do testing and validation work didn’t work very well.  They missed some relatively obvious defects introduced during the updates, both in the user interface and the business logic of the application.  It must be noted though, that they weren’t pushed very hard or given a lot of time for this work.  Given more time and direction, they may have been more successful.  Last, they may have had difficulties with testing and validation work because they didn’t fully understand the applications.  Further training would have helped with this.

3.3.    Communication

The contractors were provided with a class and methods that encapsulated most of the changes required.  This greatly simplified their training, development workflow and effort.  It also minimized their changes, which in turn reduced the possibility of introducing bugs, straying from the design pattern, and our source code review effort.
A design document was also provided to the contractors, clearly detailing their aspect of the project, development environment and their work flow.  Examples of all the source code that required updating was provided in the document.  However, the design document alone was not enough.  A screen sharing session with a Skype conference call was used to step through the workflow they were to follow.  The screen sharing session was definitely required, and was likely more effective than the design document.
As it is with managing any inexperienced or unfamiliar resource, it was important to frequently follow up with them, ensure that they are working on tasks, that they were keeping busy and hadn’t hit a roadblock.  Often roadblocks were easily solved by a second set of eyes, and resources tended to avoid asking for help or advice and remain unproductive.  It was made clear to contact us whenever running into issues and/or if they had nothing to work on.  This was fundamental to keeping the developers productive.  Note that it is still important to regularly check up on them because sometimes they don’t realize the problem, or continue working without realizing they are astray.
It is important to provide all possible contact information, and the situation to use each communication path.  Several methods of communication were provided, including phone, Skype calling, Skype IM, email and comments on work items in RTC.  Each has advantages and disadvantages, and it’s important to use them effectively and for the contractors to feel that you’re easily accessible.
At times the contractors chose the wrong method of communication.  Therefore, it’s important to correct them ASAP and clearly define the method of communication that is appropriate for each situation or problem.  For example, adding a comment to a work item is great and highly recommended for documentation purposes.  However, if the contractor is blocked and can’t continue working until the comment is resolved, adding a comment to a work item is not adequate.  Feedback on the comment could be delayed, leaving the contractor waiting and unproductive.  Therefore, some other form of communication that guarantees immediate feedback should also be used.

3.4.    Online Project Management Software

Online project management software is essential when working with remote contractors.  To properly manage resources and keep them productive, documenting, assigning and queuing tasks is required.  Documenting the tasks reduces the need for direct communication, and queuing tasks ensures there is a backlog of work to keep resources busy.  Both are important because there is a reduced number of overlapping work hours that allow direct communication regarding issues and new tasks.  Maintaining a queue of tasks keeps developers busy even when there is not the time to assign more work due to higher priorities and distractions.
Repositories of tasks that include details and estimated effort, are assigned to developers and are regularly updated also allow managers to monitor progress and productivity on their own time without interfering with developers.  It is also a relatively good documentation tool.
It is good practice to maintain a queue of tasks for resources to work on so that if they hit a roadblock and others are not available to help remove the roadblock, developers will not run out of tasks to work on.  They can document the roadblock and request help, then move onto another task while waiting for feedback.  If appropriate, it also allows developers to select tasks from a shortlist of equal priority tasks.  This gives them a bit of variety and empowerment, helping to keep them interested in the project and work when they need a change of pace.
When using online project management software, it is imperative that all resources involved in the project use the repository and keep it up to date.  Otherwise the repository becomes dated and useless, and can’t be used for documentation purposes.
The online project management software used for the project was RTC.  It was more than adequate for the project, and provides several features that were not exploited.  It was only used for source control and to create and manage work items.  Arguably, it was overkill for the size and requirements of the project, and at times seemed slow and cumbersome.
Some alternatives to RTC are Redmine, Trac and Trello.  Redmine and Trac are both open source free alternatives that can easily be self-hosted.  They provide the ability to manage and schedule tasks, as well as integration into external source code repositories.  Trello is a hosted solution that seems to be well suited to Agile development.  The interface is very intuitive and easy to use, although it does lack effort tracking, and general reporting and status features.  Last, since Trello is not internally hosted, before using it you should review their privacy statement and avoid including sensitive information.

4.     Conclusion and Recommendations

There is a lot of potential in reducing costs by outsourcing development to remote contractors.  However, overlooking the added management and learning curves could be costly.  For the project involved, the development outsourced was primarily focused on updating an existing application, which involved relatively simple work flows and the development was highly repetitive.  By doing so, management, learning curves and code reviews were minimized.  The contractors were involved with a very limited amount of development that was more technically challenging, so it’s difficult to determine how effective outsourcing this type of development would be.
Due to latency issues, it should be assumed that remote contractors must develop using local environments.  The communications latency present when contractors access remote environments is far too great to be productive.  So that a project utilizing outsourcing starts more smoothly in the future, a complete development environment should be created, verified, packaged and delivered to the contractors prior to their starting date.  Development workflows and expectations should be well documented and provided to the contractors as well.  Furthermore, these should be reviewed with each contractor via a phone or Skype call, and a screen sharing session.
Online project management software is a necessity due to the geographic distribution of the project team and the limited overlap of business hours.  It allows assigning tasks to developers, creating backlogs of tasks to keep developers busy during non-overlapping business hours, along with the ability to clearly define implementation details.  This allows managers to keep resources tasked throughout the project.

2013-01-18

Tips to Maximize Productivity - Cygwin



Cygwin is a project of a collection of tools that provides a Linux console based environment to Windows.  It is free software.
When internationalizing an application, matching keys in the source code with lines in properties files is a relatively large task.  Going through the entire application for each language, and ensuring all properties referenced in the source code are also listed in the property files is time consuming.
An even more quick and efficient method of finding missing keys is to write scripts to parse the source code.  Scripts that parse the source code for keys and match the keys with keys located in the property files can be written using Cygwin and some creativity.  Cygwin provides a command prompt that gives access to a collection of common Linux/UNIX tools.  By employing tools including find, xargs, cat, grep, sed, wc, sort and diff, mismatched keys can easily be found.  This includes keys either located in the source code or property file and not in the other, as well as duplicate keys.
These tools can be used for other tasks as well:

  • Searching projects for strings
  • Counting instances of specific strings for estimating effort
  • Listing files that require translation so that they can be added to tasks/work items
    • Calculating line counts of the files to estimate effort
  • Generating repetitive script files that apply changes to a database
  • Mass renaming of files
An example of one of a script that I created, full of relatively complicated commands to get keys used within properties files is the following, where getEscapedHTMLString and getEscapedJavaScriptString are methods that are called every time the properties file is accessed.  It took several attempts to get correct and was built slowly piece-by-piece.  The result is a sorted list of all the property keys referenced in code.java.  I added echos throughout to display progress.

#!/bin/bash
# copy Java files
echo Cleaning local directory...
rm -f *.java .properties PROPERTY_* JAVA
echo Copying Java source files...
find /cygdrive/c/src/ | grep "java" | xargs -I file cp file .
rm Internationalizer.java
# copy properties files
echo Copying properties files...
cp /cygdrive/c/src/XXX*.properties .
# get all property entries from Java source into file JAVA
echo Parsing Java source files...
cat *.java | sed "s/in\.get/\nin\.get/g;s/\")/\")\n/g" | grep "getEscape\|getPlain" | sed "s/in.getEscapedHTMLString//g;s/in.getEscapedJavaScriptString//g;s/in.getPlainString//g;s/(\"//g;s/\"[ ,a-zA-Z0-9)]*)[\;]*//g"  | sort > JAVA
# get all property entries from property files into file PROPERTY
echo Parsing properties files...
cat PIOS_en_CA.properties | sed "s/=.*//g" | sort > PROPERTY_EN
# Output properties missing in properties file
echo Missing English Properties \(PIOS_en_CA.properties\)
diff --ignore-blank-lines --ignore-all-space JAVA PROPERTY_EN | grep "<"
echo \*\* PROPERTY KEYS THAT ARE USED MULTIPLE TIMES IN JAVA SOURCE WILL BE REPORTED \*\*
echo Removing copied files...
rm -f *.java .properties PROPERTY_* JAVA

I used this script frequently to match keys referenced in the source code with property file keys.  This ensured I wouldn't get exceptions during execution, and quickly scanned for common typos.
It's also really useful to estimate effort.  In the same project, translation effort was estimated by the number of lines of code that needed internationalization.  I took the complete list of files and broke them down into functional categories to make testing/verification easier.  Then, to quickly determine the number of lines (which was later converted to effort in hours), a ran the following which outputs the total number of lines:

cat file1.java file2.java file3.java file4.java | wc -l

The last, and another relatively simple example is breaking a massive log file into pieces.  A client didn't have rotating logs implemented on their server, which resulted in a 2GB log file.  Obviously the log file is much to large to open and analyze as is.  However, by using split it's really easy to break the file into multiple manageable files.
The nice thing is, all of this is command-line based, so it's resource efficient and generally won't tie up your computer.  It's also extremely easy to save bash scripts for reference or repeating execution.

2012-12-21

Tips to Maximize Productivity - Sikuli



Sikuli is an open-source research project developed at MIT and is released under the MIT License.  It can be used for countless tasks, and scripts can be exported to an executable to simplify execution.  It is also Windows, Linux and Mac OS X compatible.
Sikuli is a technology that allows you to automate interacting with graphical user interfaces (GUI) using screenshots and keystrokes.  Developers can quickly write scripts, written in Python, which interact with any GUI including applications and web pages.  Sikuli is a heavier utility than AutoHotKey.  However it is much more powerful and serves a different purpose.  The feature that distinguishes Sikuli is interacting with GUIs based on screenshots.  Partial screenshots are added directly to your script as arguments to methods including click, double-click, find and exists.  When executed, the desired GUI interaction is achieved using image pattern matching with the screen.
Another nice thing about Sikuli is that it is an automation tool with scripts written in a full-blown programming language, Python.  This allows you to write complex scripts that support interacting in a flexible and adaptable manner based on external and unknown dependencies.  Scripts can interact with error messages, log results, and iterate a number of times that is unknown during programming and based on GUI elements.
Here are some examples of tasks that can be automated with Sikuli (website and/or application):

  • Boundary testing
  • Exhaustively testing user options
  • User Acceptance Testing (UAT)
  • Regression testing
  • Applying a base load
  • Automatically setting user preferences, changing configuration, opening applications, etc… when logging into a computer
  • Automate repetitive steps or tasks during development or debugging
    • Ex. Logging in and navigating to a specific area of a website or application
  • Any repetitive task