Take advantage of GCP Tau T2A instances during public preview
In this pun-filled article, I aim to provide a better understanding of ARM and explain how switching over can potentially improve your workloads’ performance.
Introduction
Tau VMs introduced in mid-2021 are Google Cloud’s family of VMs that offered superior price-performance ratio powered by AMD’s EPYC processors. Recently, they added a new member to their Tau family of VMs in the form ARM-based processors called the Tau T2A.
ARM is based on the Reduced Instruction Set Computing (RISC /risk/) architecture with the simple underlying concept: reduce the instruction set, but have them highly optimized. This is in contrast to Complex Instruction Set Computing (CISC /’sisk/) which is the basis for the AMD and Intel x86 CPUs.
Shown below is the CPU performance equation:
CISC architecture’s philosophy towards improving performance is to reduce the number of instructions per program at the expense of cycles per instruction, whereas RISC architecture aims to reduce the number of cycles per instruction while running more instructions. As a simple example, if you think of a “U-turn” as a CISC instruction, the RISC equivalent might be “turn left, turn left”.
Increasingly RISCy
ARM architecture got its start in mobile devices (from as early as 2002) due to its low power consumption. If you have ever owned a BlackBerry, iPhone or Android phone, it was probably powered by an ARM-based processor.
In 2012, the first Raspberry Pi was released. It is a single-board computer with the dimensions of a credit card, powered by an ARM CPU (as a side note, I myself use Raspberry Pis in my home lab). The Pi’s accessibility further ushered ARM-based computing into everyday home use.
A more recent example of ARM gaining traction is when Apple introduced their own ARM-based M1 processors to replace the Intel-based ones that ran inside their Macs and MacBooks. The general consensus is that they perform better and battery life on the MacBooks have increased drastically.
RISCy business
The Tau T2A, ARM-based VM instances have now made their way to GCP, so what does that mean for you and how can you benefit? Truth is not all workloads will benefit from ARM-based processors and any code or containers you are currently running will have to be recompiled or built (but likely with little to no code changes). If you are using base images/containers provided by a 3rd party that do not provide an ARM-compatible option, you may wish to build your own, but first verify that doing so will not void any support agreements.
Here are some examples of workloads that are generally quite suitable for ARM-based VMs:
- web servers
- embedded systems development
- Android development
- video & image processing
NOTE: you should be aware of these additional limitations
Try T2A RISC-free
Available in select regions during the public preview, the T2As are free to use so this would be a great opportunity to try them out! Complete some performance benchmarking and see if your workloads can benefit from running on the ARM architecture.
In addition, because GKE, Batch and Dataflow workloads all run on VMs, you will have the option of running these on the T2A machine types as well.
NOTE: if you do decide to take the T2As for a spin, please remember to clean up any resources you created during testing before it becomes generally available (GA) in a few months.
ARMs race
Moore’s Law has started to hit a wall (such a bold projection was unrealistic to begin with) so do not expect to see too much growth in x86 processors in the upcoming years. ARM-based processors on the other hand, do not require as many transistors. While they are reliant on memory, which used to be a major roadblock for RISC architecture, prices have come down drastically in recent years. I think we are witnessing the inflection point for ARM architecture in computing. Apple certainly thinks so and in the last few years we have come to see major cloud providers also start to offer ARM-based options. I think you can expect to see more ARM-based solutions and offerings in the future.