We reached a point where the business was under pressure. We needed more capacity, but we had two realistic options: build another recording studio or raise our prices. Neither felt right. Building is expensive and slow; raising prices can alienate the very artists who keep a studio alive. Instead of looking at the walls of our rooms, we looked at the systems between them. We found that our capacity problem wasn’t a lack of square footage—it was a storage bottleneck that was costing us time and flexibility.
At the time, that bottleneck took the form of the “three-way sync.” Our setup was straightforward but dangerous: each studio had its own system hard drive and a multi-terabyte recording hard drive. At the end of every session, we had to run synchronization software (GoodSync, if memory serves) to keep all three studios aligned. It was effectively a three-way sync, and in practice, that meant it would screw up regularly. When it failed, we had to spend precious time on recovery rather than creative work.
The real cost, however, wasn’t just in the risk of data loss—it was in the human schedule. We operated with a thirty-minute buffer between sessions. But because the syncs could take fifteen minutes or more and required an engineer to monitor them closely to resolve conflicts, that buffer was a fiction.
Our engineers were sacrificing their bathroom breaks and their meals just to ensure the next session could start on time.
When your team is monitoring a sync bar instead of resting, they aren’t just tired; they’re less effective. A more rested employee results in a better session, happier artists, and more repeat bookings. The technical bottleneck was actively degrading our culture and our product. We needed to stop chasing the sync and move to a single gold copy that was backed up and available everywhere.
This wasn’t easy to implement. This was around 2013, back in the dark days of internet cloud drives, long before fiber connections allowed you to push massive amounts of data offsite in real time. We had to run a careful hybrid cloud setup, perpetually balancing our in-house sessions against the slow crawl of backing that data offsite on a 300 megabit internet connection. It was constantly playing catch-up.
A single gold copy
The immediate challenge was serving three active recording studios simultaneously without creating a new bottleneck. At the time, our version of Pro Tools didn’t have RAM caching for audio. Every time someone hit play, the storage array had to serve multiple rooms at once across our internal gigabit network. If we simply centralized the data, we risked a system that would choke the moment multiple rooms demanded high-bitrate audio.
The foundation was a four-drive RAID 5 array using spinning hard disks of two terabytes each. But to get the most out of that, I introduced a layer of caching. I used PrimoCache on Windows Server to leverage a few gigabytes of extra RAM and one of those newfangled solid-state drives (256 gigabytes, as I recall) to create an additional cache. The goal was to ensure that whether we were recording or playing back, the system wouldn’t flinch.
Testing beyond the working load
I didn’t want to guess if it would work; I wanted to know exactly where it would break. My acceptance test was simple: the system had to be able to saturate the gigabit cable for 15 minutes without running out of steam.
To test this, I staged a worst-case scenario. I initiated simultaneous laughably large Pro Tools imports in two studios while recording 24 or more tracks in the third. We sustained that load for 15 full minutes. This was probably two and a half times more data than we would ever actually see in normal practice, but that was the point. I believe in overbuilding, underpromising, and overdelivering. Nothing went live for my employees or our artists until I had pushed the system harder than I knew the studio ever could.
Once the system passed that soak test and went live, the shift was immediate. The fifteen-minute sync battles disappeared, and the buffer between sessions became actual downtime. That relief of pressure flowed directly into the quality of our sessions. The artists were happier because the engineers were more present and less frazzled.
More room without another studio
We also saw a measurable spike in our business efficiency. We track our “booking efficiency” as a concrete measure: completed booked hours divided by technically available studio hours across the three rooms. We tracked that number month after month, watching it move from a little under 70% to approximately 82–83%. In my experience, 70–75% is a typical ceiling for appointment-based businesses; breaking past that showed us that we hadn’t just given our employees breaks back—we had expanded our capacity.
Because the data was now centralized, artists were no longer locked to a single recording studio. If someone needed a last-minute session to fix an adlib before a release or a quick radio edit, we could slot them into any available room.
This flexibility allowed us to think differently about our service design. We were able to run a separate edit suite as a “fourth studio.” Because it had access to the shared storage, we could complete simple tweaks and changes there without burning a full recording studio hour. This meant we could offer lower prices for simple edits while keeping our primary rooms available for high-value recording sessions. We had increased our capacity without building new walls or raising our prices.
Looking back, this project reinforced a principle that guides everything I do now:
Good system design should enable more human time for work—either more work or better work.
Whether it’s a storage array or a business workflow, the goal isn’t the technology itself; it’s the capacity it creates for people to do what they’re best at!
It’s the same reason I’m building commonllama. The technical challenge is different—managing KV cache and memory for local AI—but the problem is identical. When a model has to re-read the same instructions every time a task changes, it’s like our engineers waiting on that three-way sync. It’s wasted energy and time. By building a system that retains memory, we’re doing more than just increasing speed; we’re lowering the barrier to entry so that anyone on modest hardware can access the knowledge they need without the system getting in the way.
And who knows? Maybe one day AI can finally grab some lunch, too!