The Springpath Data Platform is based on patented HALO architecture. HALO was built from the ground up and leverages memory, flash and disk resources across a cluster of servers running the applications to deliver an enterprise class data infrastructure.
HALO uniformly stripes all application data across all the servers in a cluster—independent of which server an application is located on. This helps applications to leverage all computing resources dedicated to Springpath Platform software. It allows for scaling both performance and storage capacity when servers are added, while eliminating IO bottlenecks in a single server.
Both Read and Write data are cached in SSDs, delivering the highest possible performance. Incoming writes are acknowledged as persisted after the writes have been replicated to SSDs on other servers, protecting the written data from any hardware failures. The writes are later de-staged to the persistence tier based on inexpensive hard disk drives. Hot data sets that are frequently or recently read are cached in both the SSD as well as DRAM. Flash performance combined with low-cost capacity enables the best possible cost for storing and retrieving application data at full speed.
The HALO architecture separates the persistence tier from the caching tier. This unique property allows for scaling IO performance independent of scaling data capacity. A flash heavy server tier could front a hard disk heavy server tier and the two tiers can be scaled independently based on performance or capacity needs.
HALO provides fine-grained inline de-duplication and variable block inline compression. De-duplication is leveraged across all media—memory, flash and hard disk. Unlike other solutions on the market, the sequential nature of the data layout on storage media allows for variable sized compressed blocks to be written inline and retrieved quickly, without impacting performance.
The HALO architecture provides enterprise grade data services such as snapshots and clones natively. These space efficient data services are not only fine grained, but can scale to extremely large numbers. The services allow for fast data recovery or rapid application deployment.
HALO’s Log structured Distributed Object Store layer packs data into smaller objects that are laid out in a sequential manner across a pool of servers. This sequential layout increases flash endurance and maximizes performance. The objects are also replicated in this layer for redundancy. The combination of replication and log-based sequential layout enables fast recovery from drive and server failures. This is also the layer that sits close to the pool of commodity servers on which the entire HALO stack and Springpath Data Platform runs.