Multiple classes of storage make up public clouds (Amazon and Google both have at least three tiers). One of those is made from memory like flash or even DRAM to provide high performance storage to compute intensive applications. Private clouds have the same need for high performance storage, potentially more so, since they are supporting both new modern apps like Splunk, Spark, Hadoop and Cassandra as well as legacy applications like Oracle and SQL.
Traditional data centers and cloud service providers certainly have access to traditional storage offerings like all-flash and hybrid arrays. And some of these vendors are providing a scale-out design that makes them look more similar to a private cloud offering. The problem is that both types are expensive, inflexible and lock you into a single hardware vendor, even though all of them claim to focus on software.
It may be time for data centers and cloud service providers to think differently about storage and model their internal efforts similar to those of the major public cloud providers. Despite the heated competition between them, the major public cloud provider’s storage infrastructures share some important commonalities. First, they are all truly software; configured to run on white box servers purchased separately from the storage software. There is no hardware lock-in. Second, they are all scale-out in design so the performance and capacity of the infrastructure expands with demand. Third, from a customer perspective, they are licensed by capacity on a recurring basis not an upfront cost.
StorPool’s solution is a high performance storage software for data centers and cloud service providers looking to design an architecture similar to Amazon/Google’s but internally, for their organization’s private use. It is a pure software solution, meaning the organization can use practically any server and any storage media. It is a scale-out design, which means it clusters multiple servers together, aggregating the internal storage resources, presenting a single virtual storage pool. Expansion of performance and capacity is accomplished by adding another server to the cluster.
The software is licensed by capacity on either a monthly recurring or one or three year license. The capacity charge is based on the usable capacity assigned to the StorPool software. But since adding drives to the cluster is easy, customers can implement fully built-out storage nodes and then add drives to the cluster as capacity demands warrant. The software will automatically re-balance data as IT adds new drives to the cluster.
StorPool delivers performance that rivals it name brand competitors and it does so efficiently. In one test, StorPool claimed 100,000 of random read IOPS using only three cores of each node’s CPU and only 16GB of each node’s RAM.
The software can also be used for capacity centric configurations. StorPool has a hard disk drive only licensing scheme for customers wanting to use StorPool as “cheap and deep” storage.
StorPool can be a dedicated storage architecture or given the above efficiency, can be used in a hyperconverged configuration. The hyperconverged environment would need to be based on a Linux hypervisor, but for many cloud service providers that is an ideal configuration.
Protection from a media or node failure is done via 2X/3X replication, selected by the user. While many will clamor for erasure coding, StorPool believes that for high-performance systems full 2X/3X replication is the most reasonable redundancy scheme. It is very kind to CPU resources, and provides the best performance during failed conditions. The only real negative is the cost of the additional capacity it consumes. But given that StorPool runs on truly commodity hardware and only charges for capacity used, 3X replication is more affordable. A particularly interesting option is to run the software in a hybrid mode, where the primary copy of data is on flash and the replicated copy of data is on hard disk drives.
An alternative to moving to the cloud is to learn from cloud providers and use their techniques on-premises. For medium to large organizations, given the right tools, it should still be less expensive to run internally than in the cloud. To smaller public cloud providers it is a way to achieve speed, reliability and functionality. StorPool brings those cloud tools to IT and allows them to build a high performance cloud storage infrastructure.