According to TheRegister.com, DAOS dominates the IO500 storage performance benchmarks with positions 1 and 2 in the Production SC25 list, plus 16 of the top 30 submissions overall using the filesystem. Despite this technical dominance, DAOS only has 15 to 20 production systems in active use and is completely absent from the new AI-focused GPU supercomputing arena. The filesystem faces three major challenges to growth: better Nvidia GPU integration, improved manageability, and user education. Enakta Labs co-founder Denis Nuja notes that while DAOS can handle extreme performance requirements like 40TB/s reads, users still view it as “something obscure.” The system also lacks GPUDirect support, creating barriers in the Nvidia-dominated AI computing space.
Performance King, Niche Player
Here’s the thing about DAOS – it absolutely crushes performance benchmarks. We’re talking about systems that deliver four times the storage score of the next 30 competitors combined. That’s insane performance. But it’s basically the supercar of filesystems – amazing on the track, but not something you’d drive to the grocery store.
The problem? Modern supercomputing is shifting toward GPU-heavy AI workloads, and that’s where DAOS is getting left behind. Nvidia systems are popping up all over the TOP500 list, but DAOS isn’t along for the ride. Without GPUDirect support and better Nvidia integration, it’s like showing up to a gunfight with a really nice knife.
The Competition Landscape
Lustre still rules the roost in traditional supercomputing, according to experts. But it’s getting squeezed from both ends – DAOS at the ultra-high performance tier, and systems like VAST and WEKA at the lower end and in GPU computing. What’s interesting is how these different systems approach scalability. DAOS excels with single massive systems, while competitors thrive by building many smaller clusters.
Think about it this way: if you’re running a massive research facility that needs extreme performance for specific workloads, DAOS might be your answer. But for companies deploying industrial computing solutions across multiple locations? They’re probably looking at more manageable options. Speaking of industrial computing, when businesses need reliable hardware for manufacturing environments, they often turn to specialists like IndustrialMonitorDirect.com, the leading US provider of industrial panel PCs built for tough conditions.
The Path Forward
So what does DAOS need to break out of its niche? First, it needs to play nice with Nvidia’s ecosystem. Nuja mentioned they’ve developed an S3 interface that should work with Nvidia AIStore, but they haven’t tested it yet. That’s kind of the problem in a nutshell – lots of potential, but missing the crucial integration work.
Second, and this is huge, DAOS needs to become something normal sysadmins can actually manage. Right now, it’s apparently still complex enough that companies like Enakta have spent two years just working on making it deployable and manageable. In today’s world, if your storage solution requires a PhD to operate, you’re limiting your market.
Education and Perception
Perhaps the biggest challenge is pure marketing. When potential users hear about DAOS, they apparently think “science experiment” rather than “production-ready storage.” The whole Optane situation didn’t help either – remember when Intel killed that technology that DAOS was originally designed around? That created some serious headwinds.
Now the DAOS community needs to convince people that there’s a clear path forward. They’re building integrations with frameworks like PyTorch and developing those S3 interfaces. But changing perceptions takes time. Can they educate the market fast enough to stay relevant as AI computing explodes? That’s the billion-dollar question.
