ARM is showing up in all kinds of personal compute, but it also shows promise in the data center as well.
ZDNet’s Jason Hiner recently wrote how the ARM-based Raspberry Pi represents the future of compute. ARM processors, in general, are showing up in the data center in the form of CPUs for sensors and network devices. But, what about the Raspberry Pi itself? Where does the Raspberry Pi fit into enterprise compute?
Raspberry Pi limitations
For low-demand requirements, ARM-based servers make for an interesting CPU option. ARM is both cheap and low powered compared to Intel x86-based processors. HPE has been the most prominent mainstream vendor to embrace ARM in the data center, and HPE’s ARM-based Moonshot cartridges represent the state of the art of ARM data center computing. However, the Moonshot platform is more similar to traditional servers than the Raspberry Pi. Moonshot offers many of the classic features of a server platform, while Raspberry Pi represents a complete departure.
The Raspberry Pi’s architecture is similar to that of a smartphone with little to no redundancy. It’s basically a single board with integrated storage, network, and compute. The system offers USB to extend I/O beyond the integrated components.
The single board design of the Raspberry Pi limits the I/O potential of the platform. But, the Raspberry Pi is a popular platform to develop both apps and new IoT devices without the sizable investment in custom hardware. With the low startup costs comes severe limitations for moving data, which is a key attribute in most data center applications. For example, TechField Day Organizer Stephen Foskett executed some performance tests on the Raspberry Pi 2 (Pi2), clocking the Pi2 at 95 Mbps in network transfer and 25 MB/s for storage transfer. Most desktop PCs are able to push close to 1Gbps in network bandwidth and 150 MB/s in SSD-based storage transfer.
Potential data center applications
With limited CPU and I/O capability, what exactly does Raspberry Pi offer? Raspberry Pi offers cheap and deep compute. There is no shortage of examples of Raspberry Pi clusters. One of the more interesting use cases I ran across is the ability to build an object store using a cluster of Raspberry Pi Zero.
The use case for object storage is growing in the enterprise. Solutions exist on the market to present object storage as NFS or SMB. Object storage is also a popular backend for disk-based backup, and digital-focused businesses are writing applications that natively use object storage. As it further develops, a platform such as Raspberry Pi potentially makes sense for a non-performant layer of storage such as object stores.
Startup OpenIO provides a step-by-step guide for deploying their software-defined object storage platform on a Raspberry Pi Zero with only 1 CPU Core, 512MB RAM. The resulting cluster is a full-featured, S3 compatible object store that provides the ability to test S3 compatible use cases including cloud-native applications and backups.
Beyond storage are general purpose compute use cases such as edge computing. Raspberry Pi deployed at the edge of the network could monitor IoT sensors and perform lightweight scripts such as a shutdown of a malfunctioning system, or sending a notification to an operations center for further investigation.
What the Raspberry Pi lacks in pure power and performance, it makes up for in the number of nodes it offers for the price. Administrators willing to get their hands a little dirty with open source and Python have the ability to extend the capability of these tiny devices, while peering into the future of data center compute.