Attala Intelligent Storage Nodes
Attala intelligent storage nodes belong to a family of Attala products used to create composable cloud infrastructure and multi-tenant hot data lake solutions for workloads running on X86 or GPU compute nodes. These storage nodes use industry-standard NVMe and Ethernet protocols to interoperate with various host adapter options. All nodes include an intelligent FPGA-based controller that acts as the system controller to create a server-less, high-density and low power solutions. One or more of these nodes act as a scale-out cluster of elastic NVMe volumes which can be arbitrarily allocated to host servers, VMs or containers. The node includes full management functionality for the data node including discovery, chassis, SSD and fabric management and monitoring/telemetry.
DN3240 and DN3250 Intelligent Storage Nodes
32 U.2 or Ruler NVMe SSDs - single port SSDs
1U Rack high
- Four 40Gb/s or 50Gb/s Ethernet ports
- Redundant network paths to single-port SSDs
- Redundant and hot-swappable fans, power supplies
- Also available as a Supermicro product
DN2440 and DN2450 Intelligent Storage Nodes
24 U.2 NVMe SSDs - dual ported NVMe SSDs
2U Rack high
- Up to eight 50Gb/s Ethernet ports
- Dual, fully redundant IO modules to dual-port SSDs
- Redundant, hot-swappable fans, power supplies
NVMe Performance & QoS
The storage nodes deliver latency and IOPS performance nearly identical to locally-installed NVMe SSD with the industry’s lowest latency for networked storage – adding only 5 microseconds of additional latency (vs local NVMe SSDs). To make full use of the available bandwidth, any port can access any SSD within the storage nodes. And to maintain this performance given multiple contending hosts, the nodes include granular QOS controls (IOPS, throughput) on a per-SSD or per-volume basis.
The storage nodes are fully standards based & interoperable using standard Ethernet, RoCEv2 and NVMe-over-Fabrics for network interfaces; PCIe Gen3 and NVMe for its NVMe SSD interface using standard NVMe U.2 SSDs; and Redfish, Swordfish for management interfaces.
The internal architecture of the storage nodes are built on a hybrid architecture consisting of FPGA-based programmable hardware and firmware running on an embedded ARM core. This enables both high-performance and complete flexibility in how NVMe SSDs are accessed.
• For efficiency, cloud tenants or operators can carve SSDs into multiple elastic volumes with each namespace allocated to a different host
• Cloud operators can execute full SSD diagnostics (e.g. SMART) possible without user/tenant involvement
• Operators can define policies that define allowable tenant/user SSD operations
• Isolation of faults and error handling between the SSD/PCIe & Ethernet/network domains
• Extensibility for the future taking advantage of programmable hardware and firmware