Tales from the Field – July ‘21
“The Service Provider”
A few years ago, we received a call from a service provider who was stuck in a quandary. They’d been hosting a very large and intensive SQL application for one of their customers and it was time for a refresh. The issue was the cost to replace the hardware running the massive VMs required to run this customer’s VMs. Each virtual machine required 48 physical cores plus 3TB of memory, which essentially transforms each of their physical servers into 1 or 2 virtual machines. They’d invested heavily in some very large and very expensive Cisco UCS servers but we now looking for a more flexible and easier to manage design. Managing those massive quad-core servers, plus fabric interconnects, plus remote storage, plus the fiber channel infrastructure, plus the ethernet fabric, plus plus plus . . . .was difficult and, from them, admittedly antiquated.
The very idea of software-defined storage was intriguing to them due to the simplicity of the solution: servers providing storage to other servers across an already established ethernet fabric. Simple.
When the Dell team was first engaged, we started by tackling their physical storage needs. The customer’s existing storage design was comprised of multiple tiers of storage. Their customer’s “accelerated” storage was separated from the “workload” slower storage by legacy storage tiering. This allowed the service provider to charge $.75/GB/month for the fast or “accelerated” storage package as opposed to the slower “workload” storage at $.25/GB/month. We initially thought about a PowerFlex solution due to the need to grow compute and/or storage independently. But how can you tier storage in a PowerFlex solution if EVERYTHING is fast?
It’s simple really. We explained the concept of Medium Granularity versus Fine Granularity pools and the ability to utilize compression. This allowed the service provider to create an “accelerated” pool running MG pools with better performance and higher IOPs at a higher monthly rate for customer high-end workloads. They could also create a secondary tier of slightly less performing storage running another “workload” pool with FG pools utilizing compression. The FG pool also allowed for better ROI numbers for the service provider as they killed 2 birds with one stone: they had a slightly less performing tier to sell for a lower price, plus they were gaining the economies of the data reduction capabilities with PowerFlex compression. They now had their 2 separate tiers of storage – all running much faster than their previous solution!
So what about those massive SQL VMs? When talking about PowerFlex compute, never forget Dell is the PowerFlex hardware OEM with tight relationships with Intel. If the customer requires a certain number of cores and memory, and the required Intel chip exists, we can provide the nodes required utilized in a PowerFlex solution.
For this specific case we introduced quad-core R840 nodes running PowerFlex’s SDC software. These nodes included 112 physical cores and 6TB of memory. This extra horsepower now allowed the customer to have multiple instances of their VMs running not only on a single host but also multiple duplicate copies across other nodes. Not only was this redundancy appreciated but they weren’t burning an entire physical host for 1 VM.
In the end, the customers were exceedingly happy with the performance and the service provider was able to implement a next-generation storage solution with similar legacy tiering concepts. They were also able to shrink their operational costs down to a fraction of what they were doing with their old legacy 3-tier architecture.
End result – another happy PowerFlex customer.