Towards truer ‘computing on tap'

Earlier this month, AWS announced a new type of server instances called ‘T2’. These instances are drastically different from any other type in that they offer a very low-cost usage of a baseline level of CPU (10 to 40% of single core capacity depending on the instance size) and allow bursts upto 100% CPU based on the number ‘CPU credits’ accumulated. Each CPU credit allows for a 1 minute burst and a certain number of CPU credits are earned for every hour during which the baseline CPU is unused. This works absolutely fantastic for a great many workloads (see here for example).

This in itself is a major step forward for IaaS. No other instance type or IaaS provider has provided compute power based on measurement of actual CPU usage.

But what’s really exciting are the possiblities that this opens up. What I would like to see is an offering where-in I won’t need to worry about the number of CPU credits that my site might need to function well. The paradigm shift in IaaS will be when a provider allows for CPU usage to burst at any time (without the need for accumulated credits) and then bills customers at the end of the month on the basis of the number of CPU bursts / cycles used. When this happens, IaaS would have truly arrived, resonating perfectly with the long touted references of cloud to utility services like water & electricity.

All hail T2! Can’t wait for T3!!