Monday, July 16, 2012

Automation Is On the Path to Cloud

It's interesting to me how much the manufacturing world depends on automation today yet how much the majority of IT leaders fear it. Take a tour of a brewery, auto assembly line, or computer assembler and you'll see significantly more machines than people.  In college working for Allen Bradley and then Eaton Cutler Hammer who specialized in  plant floor conrol systems, I was introduced early to two key elements of cloud: distributed systems and automation. Making a trailing arm for a GM "G" car required an amazing quantity of knowledge, data collection, and computing power. The liquid aluminum had to be the right temperature, pumped magnetically into the die with the proper force, the clamping force on the die had to be within a narrow band to prevent having molten aluminum squeeze out the sides, the cooling system had to drop the temperature rapidly and uniformly to ensure the compressed crystaline structure of the metal was correct and no longer dynamic, and the die had to be cooled rapidly while being lubricated to prevent the next injection of aluminum from sticking to the hardended steel. There were natural variables that impacted quality and also had to be measured but could not be controlled so easily: humidty, ambient temperature, seismic events.  The challenge was to sense it all and manage it effectively to create parts within tight tolerances in the thousands per day.  Everthing was automated to the push of one button.

Automation is an extremly important step in moving from mass virtualization to a cloud infrastructure as a service.  At the scale and speed of cloud manual tasks just are not an option. It is amazing how many tools offer automation capabilities and how few companies take advantage. I know of story after story after story where companies were unable to meet demand spikes, traffic spikes, and even threats becuase they had always relied on human intervention but hit that out of band situation where there weren't enough humans who could work quickly enough to prevent the problem.

The value of automation is rarely argued. Everyone wants it, everyone talks about it, and it's often a requirement of a solution. So why isn't it configured or turned on?

I think the answer lies in each of three areas.  First, few people document what they do.  Automating a process is, by necessity, first a process engineering task. If you can't define what happens, when, why, the inputs, the outputs, the error conditions, and the expected results then you can't automate it.  And of course you need a period of "test and adustment" where the automation is phased in under the control of a human to ensure out of bounds conditions are tested, safeties trigger appropriately, and the system always returns to a safe state.  Second, many at the technologist level are very good at protecting their jobs and have tremendous fear in letting go of anything they perceive as "value added" regardless of how much, or how little value is actually delivered.  Third there is an inherent distrust of the unknown. Comfort with automation comes with exposure and experience.  So few IT data center operations are automated that it feels out of character to automate something. Putting the list together of atuomated tasks it basically looks like something out of the 1990's: patches (sometimes), backups, spam filtering, health monitoring, and...and...well that's about the list. Sure some go further but when I say automated I mean ZERO human interaction.

The reality is that automation not only can work, it does work, and it saves money while at the same time preventing errors. The hard fact is that an automated solution does quickly only what it was told to do.  If you don't establish the right rules the results can be disasterous.  However after 20yrs of consulting I can tell you how many process mapping documents I've seen at client companies: zero (discouting ones consultants like me created).

If you re-read the fourth paragraph I belive it provides the answer. Companies need to adopt the same engineering disciplined approach that Kraft, Intel, Ford, Proctor and Gamble and a slew of other companies already use. Define what needs to be automated and then map out the processes. The impact is tremendous measured in reducing cycle time, defects, and labor input.

No comments:

Post a Comment