[Discuss] MS Buys GitHub

Peter Willis p.willis at telus.net
Thu Jun 7 17:07:11 PDT 2018



I think there will always be a market for desktop hardware of some stripe.

In science, network computing can be problematic for some applications. Staging massive data sets in temporary AWS systems requires network transport and setup time. If massive data need to be staged and processed in a fixed contractual period of time it tends to be more effective to have the data connected locally to processing.

While consumer companies would like us all to buy into their online services, the 'last mile’ still exists and will exist out in the field for some time to come. It’s the speed of the last mile that poses the greatest bottleneck to doing data science with anything larger than a Gig of numbers.

The cloud also poses issues for analysis that runs counter to the market average. This is already seen in todays cloud services. Bread and butter are made by providing services that ordinary consumers will buy. Even some scientific platforms are rather limited because assumptions are made about how large and what formats data can be.

The cloud will also cause technology development problems because hardware becomes largely monolithic and must fit into the ‘big iron’ paradigm as deployed by the big vendors. This was somewhat true for the desktop CPU platforms we have all been using for the past decade. Manufacturers had a great deal of investment on old manufacturing equipment due to the scale of the production. Loss aversion in manufacturing has slowed the development of new technology more than anything else over the last decade.

The same will apply to cloud technology. Once MS sinks a bunch of data centres in the ocean, we’re all essentially stuck with that technology until MS decides to spend the money to recover them and deploy new hardware.

There will likely be a brief period where desktop workstations once again become a specialty business much like what SUN and SGI were selling before consumer PCs became king.





More information about the Discuss mailing list