Scale Computing Tuesday modified its HC3 hyper-converged system, adding a pair of hybrid arrays that allocate NAND flash as a tunable tier of primary storage.
The Scale flash HC2150 and HC4150 node hardware is similar to the vendor's HC3 2000 and HC3 4000 models, except with additional RAM and increased processor cores (from four to eight).
Scale isn't going all-flash with the latest rollout. Each hybrid HC2150 node is equipped with a single 400 GB single-level cell NAND solid-state drive (SSD) and three 3 TB SAS hard disk drives (HDDs). Raw capacity per three-node HC2150 cluster is 12 TB, including 1.2 TB of flash storage. The HC2150 ships with 128 GB of RAM per node and can be upgraded to 256 GB of RAM per node.
Each HC4150 hyper-converged node supports two 400 GB NAND drives and 2.4 TB of flash per cluster, with up to six SATA drives. The 4150 ships with 384 GB of RAM and can be upgraded to 512 GB per node.
Users can scale flash capacity in the units by upgrading to 800 GB capacity SSDs.
"We are not adding flash just for the sake of turning our system into a Ferrari," Scale Computing Founder Jason Collier said. "Our customers need to be able to determine how and when they use flash. We wanted to create a system where you could define the applications that need higher performance and then provide that performance dynamically."
Hyper-converged arrays consolidate computing, network, storage and virtualization resources in a single hardware chassis. Like most hyper-converged vendors, Scale requires customers to purchase a minimum three-node cluster for failover capabilities. The Scale HyperCore operating system uses no centralized management server or RAID for consensus on network outages. Any appliance in the HyperCore architecture can administer the entire cluster, spinning up backup storage as the need arises.
Scale uses the open source KVM hypervisor as its underlying virtualization technology, a move aimed at small and midsized businesses (SMBs) with distributed storage. HyperCore is a bare-metal hypervisor written around the KVM kernel.
Rather than deploy it as cache, the new Scale flash nodes use SLC NAND as a storage tier. HyperCore tiering firmware automatically stores hot data blocks on SSDs and moves less frequently used blocks to disk-based archival storage. The updated HyperCore operating system lets users assign flash to individual virtual machines or virtual hard disks.
Customers adjust the level of Scale flash for each workload by sliding a tuning knob from zero to 11. Setting the level at zero directs all writes to spindled media, while tuning it to 11 pins a virtual hard drive exclusively to the flash layer.
"What we did was to build in an auto-tiering mechanism that looks at hot blocks," Scale Computing CTO Alan Conboy said. "We apply quality-of-service-like metrics down the I/O path. Our target demographic is the Windows system administrator with a small staff. We want to let him apply flash, without making it complex."
Customers running existing Scale HC3 disk-based arrays are eligible to receive a flash kit to retrofit those systems by replacing SAS drives with SSDs.
Deni Connor, founding analyst at SSG-Now in Austin, Texas, said Scale's flash rollout addresses an underserved need among SMBs.
"What we hear from SMB customers is that they want to buy flash. The economics of buying flash are good and it's almost becoming a mainstream requirement," Connor said. "There is a need for something like this in SMBs that have transactional operations or need fast ingest of data for legal compliance."
List price for a baseline Scale flash HC2150 cluster is $61,500, and $106,625 for the HC4150 cluster. The prices include HyperCore data services for remote and local snapshots, multisite virtual machine replication and failover/failback capabilities.
Hyper-convergence offers a boost in virtual environments
Should you consider a hyper-converged data center?
Scale computing integrates Mellanox switches