I’ve been hearing claims that 2019 is the year of Enterprise NVMe. Fantastic, it’s about time we are blessed to move what was once esoteric hyperscale technology to the mainstream. Pavilion Data’s CTO VR Satish is fond of saying, “our goal is to democratize low latency.” If this is indeed the year of NVMe, we are off to a great start toward democratization.
However, NVMe alone is not the answer, nor is NVMe-Over-Fabrics. Just a Bunch of Flash or JBOFs along with old-school Direct-Attached Storage (DAS) NVMe SSDs inside servers make up the vast majority of deployments. What happened to the benefits of Storage Area Networks like common management (now called composability/orchestration), snapshots, and low-overhead RAID? These were the holy grail of SAN and ushered in an era of incredible advances. But today’s NVMe-O-F systems are just All-Flash Arrays dressed up with nothing new.
Welcome to 2019. My resolution, bring thin back in! NVMe SSDs are going to democratize low latency, but at what cost? We are still using them inside servers, creating islands of storage once again. In many customers I talk to, the most expensive storage in the data center sitting at 50% or 60% utilization. Add in DAS-based RAID protection and divide the already poor utilization by 2. Ouch!
Pavilion Data System’s NVMe-Over-Fabrics Storage Platform delivers RAID-6 erasure coding with 12% overhead, along with the manageability benefits you have come to love from SANs to high-performance, low-latency NVMe. My favorite feature right now is thin provisioning in storage.
Thin provisioning augments Pavilion’s already powerful ability to define arbitrary/large volume sizes to hosts with significantly better utilization than DAS, since the Pavilion system will only allocate physical space in the system as the application requires/consumes it. This can result in physical space savings vs. DAS of up to 80% per server. No wasted capacity, no time spent messing with volume managers or file systems. Just set and forget.
One of our largest accounts is basically able to pay for racks of our storage platform by moving their under-utilized NVMe drives out of servers and into our arrays. Since we have OpenChoice, which allows the customer to dictate how they populate the storage in the array and the timing thereof, they can repurpose large investments in WD, Samsung, and Intel 2.5-inch SFF drives into an intelligent architecture where disaggregation leads to less aggravation.
I’m just getting started at Pavilion Data Systems. Yes, my first resolution is to bring thin back in, but please stay tuned as we dig deeper into this powerful NVME-O-F Storage Platform.