Hi Natalya, We fully agree with you and this is exactly why Scality launched ADI, a new storage software designed specifically to address this problem. Out of every player in the industry, Scality is the one who has most invested to deliver SSD like throughput on HDD. We have production deployments closing 422 GB/s on HDD, and TB/s in sight (with proper network). In addition we recognize that to fully benefit from each media type, from ultra fast TLC all the way down to tape, it is critical to have a single unified namespace independently of media. S-CSP, we are ready for you. This was the core of our message at RAISE Summit last week. Jerome
The #neocloud (or, as IDC calls them, S-CSPs - specialized cloud service providers) market has a #storage problem that nobody thought about when the GPU land grab was in full swing and it is now causing some headache. When the first wave of S-CSPs entered the market, storage wasn't a big focus for them. They were laser-focused on GPUs and many had specific experience in building accelerated computing infrastructure, but not storage. Storage was a fraction of their budget, and all-flash was the default. It was simple, fast, and at the time, economically defensible. That window has closed. Memory and SSD prices surged over 60% in a single quarter, while the demand for storage kept growing. So, the low storage budgets are no longer an option. And S-CSPs also face a deeper architectural problem - the lack of storage tiering. Unlike them, hyperscalers never went all-flash. They run flash and HDD storage in the same data plane with intelligent tiering, which gives them more leverage for controlling storage budgets. S-CSPs built on all-flash have no equivalent lever to pull. Storage is the part of AI infrastructure that S-CSPs underinvested in when it was cheap and are now paying for when it isn't. That sequence is rarely kind.
