During my time as an ”Ops Guy” at Electronic Arts, I became very interested in the future of private cloud. For many companies, the future of cloud is a confusing one. While public cloud can provide great benefits such as speed and flexibility, anyone with a pen and paper (or spreadsheet) can quickly see that it is not always the most cost-effective option. As a result, more and more Enterprises are turning to private cloud for some of their applications.
Before going hog wild and building our own private cloud, it’s worth taking a look at what we have learned from many years of leveraging public cloud. If public cloud has taught engineers anything, it is how to build and deploy applications on commodity hardware. Developers moved applications into the cloud fully aware that the server they are running code on might disappear at any moment. Software development of today has made leaps in bounds towards the idea that an application must be able to deal with server failure without freaking out. The days of yesterday where IT professionals cared for their servers in a way akin to a mother raising her young are gone. Servers today are viewed as ephemeral, being stood up and torn down through standard CI/CD processes. As someone recently said to me, “Server hugging is so 2000’s”
With this knowledge, the question becomes: Can we deploy an internal cloud using some of the same principles of public cloud, namely reduced cost through commodity hardware? Enter the Intel Compute Card. The Intel Compute Card was first seen at CES 2017 and was originally developed to power the next generation of appliances (TVs, Refrigerators, etc..) But for many reasons, Intel may have accidentally built the perfect cloud server.
Coming in at 94.5m x 55mm x 5mm the intel compute card is small in stature but not in capabilities. Sporting a dual-core i5 vPro Intel processor, dual-channel DDR3 4Gb memory, 128Gb SSD storage, integrated ethernet, and graphics this machine has more than enough power to handle most micro-services that are the trademark of today’s distributed architectures. Things get very interesting when one looks at the power consumption of this tiny beast. At ~20 watts, you can get almost 17x the number of cores and memory when compared to a standard 1U server (300 Watts). Furthermore, The Intel Compute Card comes in at just under $200.00 and has a unified form factor for easy replacement. Current docking devices come with an actual eject button, making swapping out a dead node very easy (Seriously – 3 ½” disk style).
One can quickly imagine a world where your application nodes are powered by disposable, unified, compute cards. No need to worry about “hardware refresh”, simply pull out the card and slide a new one in. Now obviously, this technology might not be your first option when choosing where to put your database, but as a K8s node? Interesting….