At its most basic, computational storage involves putting processing power where the data resides. In the case of flash storage, a computational storage vendor installs processing power on the flash drive. Imagine an array full of 100 flash drives; now imagine that array full of 100 flash drives each with an individual processor. When an application requests data from the array with standard drives, the drives must send all the data to the requesting application. But when an application requests data from the array with computational storage, only a sub-set is sent back to the application.
Computational storage reduces the need to build massive computing farms with high-performance networks. Computational power scales as IT adds more capacity to the infrastructure. Applications that interact with an enormous data set, to determine a much smaller answer, are ideal for the computational storage use case. Examples include real-time analytics, as well as intelligent edge, and content delivery networks, among others.
The challenge facing computational storage vendors is how to broaden the use case beyond a few highly advanced customers with the resources to perform significant internal development. Getting your application ready for computational storage is becoming easier. In some cases, the app will run unchanged on the computational processor, and in others, a developer needs to recompile it for a different class/type of CPU.
In Storage Switzerland’s panel discussion, “Is Computational Storage a Better Path to Extreme Performance?” we are joined by a panel of experts to discuss:
- What computational storage is
- Who needs it
- What it takes to get started