Enabling Scalable Flexible Deployments
RainStor’s specialized database is a massively parallel processing (MPP) system with a logically shared everything architecture. RainStor shares all of its workloads across multiple servers in a RainStor Cluster by means of Service Managers. These Service Managers ingest data, de-dupe and compress into multiple partitions, which are stored in on-disk containers and immediately made available for query, with the query workload distributed across all available nodes.
RainStor is a self-describing, zero admin repository that can leverage standard RAID or be deployed under the Hadoop File System (HDFS) to take advantage of built in redundancy and replication.
Furthermore RainStor’s massive scale compression and large block data access effectively makes it a hardware multiplier and with less disk I/O required, expensive Storage Area Network (SAN’s) are not needed. You can run RainStor on the lowest cost commodity hardware (NAS) or run on a dedicated CAS device – all of which can be run on-premise or in a hosted environment by industry solution or service providers.
As data volumes grow or the need for greater reporting access expands, additional nodes can be easily added. RainStor is therefore very CPU friendly and additional RainStor servers can be added to the cluster through simple configuration. Similarly, a server can be dynamically removed by server shutdown or by reconfiguration at the Service Manager level, effectively putting it into a dormant state.
Deployment Models:
The customer choice ultimately depends on scalability requirements and the desire to reuse existing in-house infrastructure where cost is often a key factor.
In a shared disk deployment, the data is landed in a staging area and processes grab data as it becomes available to build the RainStor partitions during the ingest phase. When a query is requested, RainStor creates a query plan, which is executed in parallel by multiple nodes accessing the shared data area.
In say a Hadoop distributed file system (HDFS) deployment, the data is landed into separate staging areas specific to each node. RainStor partitions are built in parallel by each node during the load or ingest phase. When a query is requested, RainStor creates a query plan, which is executed in parallel by multiple nodes leveraging the distributed file system to gain access to data on other nodes if needed. Because RainStor can achieve significant compression during ingest, this allows for easier shipping of data and minimal performance cost.
Highly Available:
Because RainStor is constantly load balancing and sharing tasks across nodes, failure of an individual server could lead to some degradation in performance but will not impact RainStor availability and additionally, ingestion and queries will continue to be executed.
The RainStor architecture supports backup and recovery with industry standard tools and the chosen backup schedule can include any combination of full, incremental, or differential backups according to recovery time objectives. RainStor can also support block-based replication to a secondary instance in a geographically distant location for improved levels of disaster recovery.
