Architecture

/Architecture
Architecture2018-12-05T00:10:36+00:00

The Re-invention of the Server Platform

Data-intensive workloads are dynamic and often drastically different. So, how do you plan server infrastructure that is optimized to the needs of each workload? Composable Infrastructure. At DriveScale, we re-imagined the infrastructure you need to plan, optimize and perform – all at a lower cost.

Composable Infrastructure begins by disaggregating the compute from the storage that is currently aggregated in direct-attached storage servers. With standard top-of-rack Ethernet switch performance at 10G, 25G and 100G today, all of the bandwidth needed for substantial storage traffic is available.

While SAN and NAS are networked storage technologies that are in use today, they both fall down when it comes to scale, performance and cost. The latest option, software-defined storage (SDS), is attempting to lower costs by leveraging direct-attached storage (DAS) servers, but it doesn’t scale and it doesn’t solve the problems DAS creates. SAN, NAS, and SDS aren’t viable solutions for data-intensive computing workloads.

DAS Server

Today, most data-intensive computing workloads are run on direct-attached storage, generally considered a commodity solution with very good performance. DAS, however, causes complexity and added costs from overprovisioning and merged resource life cycles since compute and disk are locked together in the configuration purchased. Since most applications and workloads require a different amount of compute and storage, server SKUs proliferate causing operational complexity. In addition, the fixed resources in DAS servers makes it impossible to manage the upgrades of compute and storage separately or to adapt the amount of compute and drives to the demands of the changing workloads. Most often DAS servers are overprovisioned and underutilized.

Composable Infrastructure brings a substantial change to how we can now define a server. With DriveScale, highly flexible and adaptable configurations can be deployed using a single compute and a single storage building block. For a latency-sensitive application, you can compose an all-flash system with exactly the amount of compute and flash you need. For infrequently accessed data, you can connect 100 disk drives (or more) to one compute node to lower costs and increase utilization. This new level of flexibility gives you the control you need over your infrastructure.

Low-Cost Servers and Storage

At DriveScale we support your choice of third-party diskless servers and dense storage systems including JBODs and eBODs (Ethernet Box of Drives) for flash or HDDs, connected to a high-performance Ethernet fabric. Our storage adapter software runs on the eBOD storage systems to bridge protocols including SAS to Ethernet and NVMe to Ethernet. The DriveScale Adapter Software provides a storage to network protocol bridging stack, disk discovery, inventory, and monitoring, load balancing and flash drive slicing.

diskless server

Diskless Server

Flash eBOD

HDD eBod

HDD eBOD

The DriveScale Composable Platform

Core to the DriveScale software platform are 4 key architectural principles: high availability, direct-attached performance and low latency, massive scale and security.

We have designed in redundancy and path management at every level. We start with dual-ported disks each connected to two controllers either inside or outside of the eBOD (Ethernet Box of Drive).  Storage adapters, eBODs/JBODs, compute nodes and network switches are all dual connected. The switches are configured in multi-chassis lag, traffic is load balanced and the paths are monitored so if anything fails the traffic is rerouted. Our patented load balancing manages traffic for congestion and our patent-pending data placement system ensures data resiliency across bandwidth and failure domains.

The DriveScale platform software is designed and tuned to ensure DAS-like performance as if the compute and storage are still in the same chassis. For the DriveScale NVMeAdapter, which is a standard flash chassis with controllers running our adapter software, you choose the third-party flash drives that meet your performance requirements. With 4 X 100GB links and 24 bays for drives, we provide 2MIOPS per adapter.

We designed a shared-nothing, parallel, compute software architecture so the servers are independent and do not compete for memory or drives. This enables our extraordinary scale of compute and storage to the size of cloud-scale clusters. Moreover, we maintain flexibility, high performance and high availability at scale.

Core to the platform is an advanced and automated encryption key manager, encrypted messaging and full drive encryption. Working with Whit Diffie, the world-renowned pioneer in cryptography, we created a comprehensive and very strong encryption solution. Crypto keys are based on a customer secret and enhanced with added security measures. All messaging between elements of the DriveScale solution are encrypted. When turned on, all data written to disk is encrypted along with any RAID, partitioning or other data. The solution leverages standard Linux dm-encrypt to encrypt the data. If a disk is taken out of a server cluster (or out of the data center), the data can’t be read. If disks are moved to new compute nodes, the keys move with them. It’s flexible, simple and safe. To learn more, get the whitepaper.

The DriveScale Composable Platform includes DriveScale Composer, the orchestration, monitoring and management application; DriveScale Server Agents for discovery, inventory and monitoring compute nodes, as well as deploying encryption; and DriveScale Adapter Software for connecting drives to the network, and the discovery, inventory and monitoring of drives.

With the DriveScale Composable Platform, you create servers from commodity elements that scale, perform and remain available. It’s fast, flexible and always on. It’s DriveScale.

Request Demo