As the lead data scientist at AiR HUB, a significant part of my job is just managing data. We get a lot of sensitive info from our industry partners, and we definitely can’t just slap that onto the public web.

This brings up a big question: where do we put it all? Cloud services are great, but ideally, we want to keep everything on-premises right here at the International Gaming Institute. The catch is that local servers usually can’t match the uptime guarantees that the big cloud giants promise.

Cloud vs local

Personally, I don’t think the cloud is worth it unless you’re hosting something public-facing. Since a fair chunk of our data isn’t for public eyes, putting it on the cloud doesn’t make much sense for us. Plus, we didn’t want to be 100% dependent on the cloud for all our computing power.

Local infrastructure is harder to maintain, sure, but it gives us the freedom to use whatever tools we want. Relying on the cloud locks you into proprietary software stacks that are hard to change, and the bills can add up fast.

And let’s be real about reliability: basically every cloud service has gone down at some point. AWS, Azure, Cloudflare, they’ve all had outages. So when providers claim “100% uptime,” it’s mostly just marketing fluff.

The main headaches with going local? Upkeep, security, and the fear of losing data.

Building a robust local server

One of our first big moves was setting up a high-performance local server right here in the offices of the International Gaming Institute. It’s a beast, primarily outfitted for machine learning and AI tasks, with plenty of storage and expansion for whatever dataset we throw at it. This server is the backbone of our research infrastructure at AiR HUB.

When picking an OS, you usually have two choices: Windows Server or Linux. We skipped Windows pretty quickly. Between the licensing fees, the restrictions, the constant worry about ransomware and reliance on buggy third-party security (remember the Crowdstrike mess?), it just wasn’t worth the hassle.

Linux was the obvious solution here. We went with Ubuntu Server because the documentation is excellent and it’s the standard for the kind of work we do. Linux systems are super efficient. They barely use any system resources and you almost never have to reboot them, even after updates. That means more power for our actual work.

Linux systems are also inherently more secure for a few reasons:

  • Permission systems are very granular and implemented at the file system level.
  • A Linux server isn’t bloated; it doesn’t even have a graphical interface (GUI) by default. This means that the attack surface (moving parts for hackers to exploit) for a Linux system is smaller compared to Windows.
  • Linux systems rely on package managers for installing pre-vetted software. It’s much cleaner than the Windows method of managing every application individually.
  • Because the code is open, the community finds and squashes bugs very quickly. With Windows, you’re stuck waiting for decades for Microsoft to patch things.

Is Linux infallible? No. But, 96% of the top one million web servers run on it for a reason: it’s easier to secure and lighter on resources.

Managing the Beast

We keep the server locked down in two main ways. First, you have to be on the UNLV internal network to even see it. If you’re on campus, great. If you’re working remotely (which happens a lot these days), you have to hop on the UNLV VPN first.

Second, we use SSH (Secure Shell) with a strict public key requirement. That means no passwords. If your public key isn’t on the server, you aren’t getting in. It makes onboarding new users a little more work, but it means we can sleep easier knowing the server is secure.

For backups, we follow the standard 3-2-1 backup strategy:

  • 3 total copies of the data.
  • 2 copies on different media (we have a dedicated internal drive just for this).
  • 1 copy stored off-site.

We use a tool called Borg for daily incremental backups, and we physically move data to an off-site location once a week. We also handle updates weekly, usually without any downtime.

Scaling Up

As AiR HUB gets bigger, our tech needs will too. AI is moving at a breakneck pace, and I’m sure we’ll need more servers or upgrades soon. But the goal was to build a tech policy that protects our assets right now, and I think we’ve built a solid foundation for the future.