One of the meetings at the Next-Gen Storage Summit (NGSS) was by the Object Storage Alliance, a group of many of the companies in this technology space offering storage systems and related products. There was a lively discussion about what the primary use cases for object storage should be, which ended without a consensus. However, there’s always agreement that object storage systems are the best solutions for large, unstructured data storage. Amplidata’s new Himalaya architecture is designed for just this use case.
Amplidata is a software-only object storage solution that’s used by service providers and large private clouds to create resilient storage infrastructures that can scale almost without limit. At NGSS, they discussed their new hyper-scale solution, the Himalaya architecture, and how their technology differs from that deployed by other object-based storage systems.
Amplidata’s AmpliStor solution leverages a scale-out architecture comprised of 1U storage and controller nodes connected by a GigE fabric.
Storage nodes hold up to 12 x 4TB hard drives and controller nodes run Amplidata’s erasure coding software, plus manage all the metadata handling functions that are part of a local or a geo-dispersed cluster, supporting a global namespace.
Recently, the company announced their new high-capacity “Himalaya” architecture in which modules called “Scaler” nodes actually sit in front of an existing AmpliStor cluster. This creates what I’d call “second degree” scale-out storage, or a “cluster of clusters” since each scaler node can connect to another entire AmpliStor cluster. With each AmpliStor cluster able to contain an Exabyte (1000 petabytes), this creates an enormous potential for capacity.
One of Amplidata’s differentiators is their “rate-less” erasure coding (EC) technology. Erasure coding is a parity-based process that enables the storage system to recreate a lost or corrupted data object by using redundant data blocks. EC is ideally suited to object-based architectures and is used by many object storage systems to replace RAID or a second copy of data.
Most erasure coding software allows the user to set the level of resiliency based on the proportion of redundant data blocks created. Rate-less EC allows the user to change the parity level of the EC scheme without affecting performance. According to Amplidata, this capability allows a company to reclaim storage capacity by reducing the parity level of a data set, with zero hardware impact – and no data migration.
Aside from reducing storage consumption by a specific percentage (certainly a big deal even at a small % of savings when we’re talking about EB-sized storage systems), rate-less EC also removes some of the risk of over provisioning or under protecting a specific data set.
Amplidata’s technology is focused on what could be called the primary use case for object-based systems – extreme capacity. Given the fact that an AmpliStor cluster can grow to an Exabyte, this new architecture allows each cluster to itself scale-out and certainly gets close to earning the label “unlimited capacity”.
But scalability is dependent on a sound data resiliency process, like erasure coding, since RAID rebuilds and second copies aren’t feasible with huge data sets. Amplidata’s rate-less EC technology seems to have provided more than just resiliency, but also a way to reduce risk and improve efficiency in the hyper-scale storage environment.