Why 1st appearance of ARM servers failed, but can succeed the second?

In a previous note, I mentioned HPE servers running the ARM platform, which after a couple of years, without much publicity, transferred to x86. While server ARM CPUs and servers were produced by many companies. Why did the major players curtail these products and why ARM cloud providers pay so much attention to it now?

By the middle of the `10s, a cloud-native application was quite formed in its architecture. Some of the solutions previously were not used with legacy applications: Redis for caching, ElasticSearch for search and cache, and message queues. Applications have evolved very much towards the web approach and horizontal scaling.

Initially, web and horizontal scaling of small application or server instances was the zone of interest to use ARM as server CPU. At that time ARM was quite a low-power CPU, but also consuming little energy. The very thing for application with low to medium loads such as web, or MapReduce (very fancy technology during the mentioned period), or even IoT processing. In general, all those applications may not load CPU for 100%, and quantity sometimes is more important than quality.

But the market, as always, decided on another. To begin with, very few of the enterprise customers needed ARM servers in their own DC while clouds are on the rise. To continue it turned out that performance is yet still too little. And, finally, the software did support the platform at the required level. While you could install and run Linux, most of the applications either did not support the platform or did not use the capabilities and features of the CPU (not clear what is worse).

As a result, the bright and beautiful future of ARM mass servers was washed out by reality. But ARM Holding, as a developer of the platform, did not care much about such trivia and stared into their bright future, which can be divided into two parts: clouds and 5G.

Amazon acquired Anapurna Labs, the developer of ARM processors, in early 2015 not for nothing. On the hyperscaler caliber switch to an own energy-efficient platform can save billions per year.

The best example of the result of this acquisition is Project Nitro. A joint hard/software solution that allowed to move virtualization and management overhead off servers to dedicated PCI-X board. Previously about 1/3 of the server was reserved for management purposes, now 100% can be sold.

Furthermore, there are many SaaS and PaaS services: DynamoDB, S3, SQS, etc. These services can be moved to a new platform. The benefit of such a move can be shown by Apple’s experience with its M1 and A14 CPUs. Both have units optimized for certain tasks. Basically, these units are a whole co-processor, but already built-in. The old idea gets a new life!

As a result, Amazon, Microsoft (which develops its ARM chip), as platform owners, get a specialized solution optimized for their needs. Just like IBM designs mainframe-optimized processors rather than using Intel’s general-purpose CPU (well, almost).

If ARM in the cloud is already the reality of today there is still a niche for the future: low-power and embedded servers for 5G, SmartNIC, and edge computing. Areas where low demanding platform and extensibility for special use cases will earn its success. With the spread of 5G and the gradual expansion of smart everything, the applications themselves will shift closer to data sources. The Internet of Things has not yet become a daily reality, but it has an intermediate stage - the “fog of things”. And this fog will become the computing power of all sensors and metering devices. Also, there will be smart machines - now there are few models with the support of M2M, but the concept is entering the market.

So Intel is not going anywhere and will not die, but rather release its ARM chip itself (again). And will work to ensure that x86 can move in the new growing market to move ARM away. In servers and PC, the ARM will remain a speed-up solution: business laptops, ultrabooks, and so on. Microsoft will provide the platform in the form of OS and basic software. But whether the initiative will get support and traction from vendors like Adobe, Corel, Autodesk who release highly demanding software - it`s a separate question that also will have a significant impact on the development of ARM as a platform for computers. The last stronghold that remained is games, but I would not be surprised if Unreal Engine in the next couple of years also will adopt this platform…

In any case, it remains only to wait for the server manufacturers to support the initiative and what will be the “answer to Chamberlain” from Intel.