Software Defined Storage (SDS) may still be a fuzzy concept in the average data center but its promised deliverables are in clear focus. Data center managers are looking for ways to rapidly improve storage provisioning times while increasing overall flexibility. Hedvig’s distributed storage platform is an SDS solution that seems to deliver on the SDS promises.
The Storage Sprawl Problem, Worse Than Ever
The challenge facing storage is the new agile data center. Because application demands can change rapidly storage administrators can’t accurately predict storage demand, for both capacity and performance. Forecasting storage needs for the next quarter, let alone for the next three years, is almost impossible. Most storage systems lack the scalability to rapidly meet these demands or adapt to new environments like Hadoop, Cassandra, OpenStack and Docker.
The result is that most organizations continue to overbuy needed performance and capacity, while still buying specific systems for each type of environment. This situation only compounds the fragmented storage infrastructure that data centers already face today. It is not uncommon for an organization to have a storage system for VDI, a storage system (or two) dedicated to virtual server infrastructures, another for physical servers supporting applications like Oracle and MS-SQL and still another for unstructured data. As these new environments like Hadoop, Cassandra, OpenStack and Docker move into the enterprise, the number of storage systems, each in their own silo, will increase.
A Distributed Systems Approach To Storage Infrastructure
Hedvig’s CEO Avinash Lakshman was responsible for building two of the core distributed computing systems at Facebook and Amazon. That work resulted in the creation of next-generation application platforms including Cassandra and Dynamo (the genesis of NoSQL). In founding Hedvig, he is leveraging this work to bring a distributed systems approach to storage infrastructure.
The Hedvig Solution
Hedvig’s solution has three software components. First is the Hedvig Storage Service. This is Hedvig software running on a commodity white box server configured with Flash, Hard drives or both that can also attach to existing storage hardware. These storage nodes are clustered via Hedvig’s distributed clustered engine, which seamlessly connects nodes, and provides a protocol access and data services engines. These engines are the heart of the Hedvig solution. Leveraging Lakshman’s experience, they feel their storage cluster should be able to scale further and be more flexible than other scale out solutions on the market.
As mentioned above the protocol access engine provides a full range of access support including iSCSI, NFS, S3, Swift and Cinder. The data services engine provides a complete set of data services including inline compression, inline deduplication and auto-tiering. It also works with the next component, the Hedvig Storage Proxy, to provide integrated server side caching. The client also presents block, file, or object to any compute environment via a VM or Docker container. The Storage Service and Storage Proxy can be installed on the same server to transform the solution into a hyper-converged architecture.
Hedvig’s solution could appeal to a wide range of organizations or more importantly a wide range of workloads within a single organization, eliminating storage silos. They feel the solution is suitable for virtualized environments (server and desktop), private cloud storage needs and as a storage/processing farm for big data analytics.
Over the next few years storage infrastructures have to meet the demands of the agile data center, which has become more responsive to the needs of the business, while eliminating storage silos instead of increasing them. But the agile data center makes it difficult to forecast what the storage demands will be in terms of capacity, performance and protocol support. Hedvig promises to address this via a flexible, software-defined architecture that can meet future demands and consolidate legacy storage resources.