The HPC industry is starting to experience a fundamental shift in the way it pays for storage. The accessibility and instant nature of cloud services are slowly changing the mindset of those who require compute clusters, making systems far more accessible for all.
However, the pay-as-you-go method of pricing is not sustainable. The challenges will come when companies realise quite how much they are spending.
Traditionally, when a large organisation procured IT services, whether hardware or software, a specific person takes charge. The company might put out a request for participation and various companies would bid on the work, all of which takes a very, very long time.
The cloud model has completely broken this practice because people can purchase IT resources simply with a credit card. Right from the word go, people were spending money on AWS instances themselves because it was a sufficiently small amount of money.
This feels a bit like the early days of mobile phones, when we paid for everything individually – 10p per text message, or thereabouts. People soon realised how much this added up, the phone companies reacted and now most of us don’t need to worry about how much we are phoning or using the internet because it is all pre-paid in some sensible way.
Watching the pennies
As the cloud market matures, this could be how we buy compute resources. It is all very well spending £10 here or £1,000 there on a credit card, but this all adds up. If you are thinking of moving a large portion of your infrastructure to AWS you really need to keep on top of the costs. It might only be small companies that care about spending £1,000-£2,000 a month right now, but even large companies start to care when its £200,000 or £1 million. There is quite an argument for making pricing much simpler.
No matter how this pricing model evolves, it is already affecting how people expect to pay for software. They think, if a company can buy hardware for so many dollars per hour, wouldn’t it be great if we paid for software in the same way? Annual licences for continuous access are being challenged.
Whether this is a good or bad thing is yet to be determined. Some people think of this as a way to cut costs and only pay for software while they are using it, while others are actually happy paying more for the software if they can put it down as an operational expense rather than a capital expense.
Operational versus capital
There has long been this war between software and hardware – large budgets are reserved for hardware as it’s a capital expense, and therefore considered safe, versus small budgets for software as it is deemed a high-risk purchase and an irritating operational expense.
Now that cloud offerings are moving everything to an operational expense, the conceptional difference is disappearing between a cloud instance running a purchased software licence compared to the hardware and software as a bundle. We may well see people spending more on software in the future because everything in their mind is software.
Whether this comes to pass or not, it would be wise for companies to keep an eye on their engineers’ credit card spend, at least in the short term.