Are you ready for the decentralized future?
Ever since the beginning days of "the cloud" I have always been skeptical of using "someone else's computer" to do sensitive work. I realize the incredible security effort companies like Amazon and Microsoft put into their systems -- commensurate to their Security Energy -- but there is still a component of centralized control that makes me wary.
I have been running my own infrastructure for personal projects for 20 years. Summer 2005 I fondly recall setting up my first public FTP server on the university network where we had…wait for it *10 Megabits* of upload speed. I mean you could practically rule the world with this. (For context, youngin's, I was coming from home DSL with ~300kbps of upload speed and just a few years since 56K phone modem was standard.)
This computer ended up with vsftpd on it serving up dvd rips of family guy episodes with the vision of being able to stream and watch them anywhere. Of course the software to enable this smoothly was still a few years away but the experience of setting up the server, authentication, file sharing was invaluable in shaping my skills to run more elaborate infrastructure.
About 10 years ago I made the leap to a "real" homelab with some surplus Dell servers running on an old rack and that has grown to now 5 servers arranged in a mixed cluster of Proxmox and Kubernetes. Two nodes, with GPUs are bare metal Kubernetes worker nodes, while the other three are a Proxmox cluster and also running VMs with about half of their resources dedicated to Kubernetes control plane duties.
The homelab journey has mirrored my professional development with the early versions being based on Windows Server and Hyper-V while I was doing a lot of Microsoft red teaming work, then later it shifted to vSphere as I focused more on VM automation and terraform, then finally to Proxmox as I got deeper into the Linux and open source world. Now Kubernetes is the mainstay as I have been immersed in devsecops the past few years building cyber ranges.
And this is where the hobby turned into something more. About 6 months ago I was made aware of the Akash network decentralized cloud provider. They have a network of providers which provide compute, memory and storage to a "super cloud" where anyone can deploy container-based workloads. There is a utility token AKT which provides the core of the network binding but they also support stable coin payments for customers wary of buying into cryptocurrency.
My Kubernetes cluster was reconfigured and added to their network as a provider earlier this year and I am now a contributing member of the super cloud. I am also starting to migrate some of my own workloads to Akash running on both mine and other providers. Depending on when you read this, the blog is hosted on Akash as well.
The primary value case for Akash is GPU renting which enables all manner of AI training and inference workloads to be spun up on demand, all over the world and I am enjoying access to large GPUs for much cheaper than other providers and away from prying eyes of the centralized providers.
I'll do more security evaluation of Akash in the future, for now I'll just share that my cluster has been more CPU-utilized than I expected. I come from a background of memory always being the bottleneck when running large simulation labs with lots of VMs. These container based workloads are so far using 50-75% of my CPU capacity while memory is sitting around 10-15%. It's paying for my homelab electricity anyways and gives me experience in running and using Kubernetes in a production capacity.
Future work will include more security audits (ideally in cooperation with Overclock labs who founded the network) as well as finally upgrading my remaining DDR3-based servers. I'm open to providing more GPU capacity also as that has the potential to make more $$$ but while there is a lot of latent capability it makes more sense to focus on applications that can use that excess…stay tuned.