Cloud PLM and Amazon S3 Scale

I'd suggest (from a position of ignorance, granted) that most of the 'heavy' data of drawings, CAD, manufacturing info etc. is at a part-type level. There'd be of the order of 3 million of these (perhaps considerably fewer, as presumably many part-types are used multiple times in an aircraft).  So 150,000 S3 objects per part-type. S3 scale is winning here.

If you want to collect data on each physical part, you're down to 200 S3 objects each. But an S3 object is potentially big (upto 5 terabytes) - way more than you'd need per physical part. So S3 again is several orders of magnitude bigger. You could even generate several hundreds of megabytes of data from monitoring each part in service and still be smaller by a distance.

And look at the growth rate on S3.

I'm not sure that product data and CAD pushes the limits of 'Big Data' as once it did. A few years ago I was told by someone at BAe Systems that the product data for the Eurofighter Typhoon was around 1.5 terabytes. They were trying to impress, but I now have 2 external hard drives on my desk that could deal with that.

Andy, Thanks for your math! Appreciate your insight. I think, we are in a very early stage of understanding how to translate the existing product data to the cloud. Your 1.5T example is good when you are talking about JPG files. Product data is different. In case of aero, the question of product serialization is interesting. Not sure it is a solved problem these days. Now, move it to the cloud, think about granularity, redundancy, access speed, and many other things. My point wasn't to say S3 is too small, but just to present a level of magnitude. A good list of unsolved problems, in my view. Best, Oleg

It's interesting to think that there have been around 4 billion parts, including fasteners, generated for the 747-400s alone. Gives some perspective over the number of parts overall knocking about, especially in the context of  some of the ideas around the Web of Things.

Not sure the issues with product data are to do with scale as such, more to do with the issues you mention: granularity, redundancy, latency, bandwidth. Deciding where to store data so that it's 'near' the user. And for CAD, do you move pixels around, or triangles or b-rep.

Andy, agree with your point. Real word is probably more complicated that web today. So, few more challenges ahead :)... Best, Oleg