Panasas ActiveStor vs Sonexion Scale-out Lustre Storage System comparison

In our most recent product, the ActiveStor Ultra, Panasas has developed a new approach called Dynamic Data Acceleration Technology. It uses a carefully balanced set of HDDs, SATA SSD, NVMe SSD, NVDIMM, and DRAM to provide a combination of excellent performance and low cost per terabyte.

• HDDs will provide high bandwidth data storage if they are never asked to store anything small and only asked to do large sequential transfers. Therefore, we only store large Component Objects on our low-cost HDDs.

• SATA SSDs provide cost-effective and highbandwidth storage as a result of not having any seek times, so that’s where we keep our small Component Objects.

• NVMe SSDs are built for very low latency accesses, so we store all our metadata in a database and keep that database on an NVMe SSD. Metadata accesses are very sensitive to latency, whether it is POSIX metadata for the files being stored or metadata for the internal operations of the OSD.

• An NVDIMM (a storage class memory device) is the lowest latency type of persistent storage device available, and we use one to store our transaction logs: user data and metadata being written by the application to the OSD, plus our internal metadata. That allows PanFS to provide very low latency commits back to the application.

• We use the DRAM in each OSD as an extremely low latency cache of the most recently read or written data and metadata.

To gain the most benefit from the SATA SSD’s performance, we try to keep the SATA SSD about 80% full. If it falls below that, we will (transparently and in the background) pick the smallest Component Objects in the HDD pool and move them to the SSD until it is about 80% full. If the SSD is too full, we will move the largest Component Objects on the SSD to the HDD pool. Every ActiveStor Ultra Storage Node performs this optimization independently and continuously. It’s easy for an ActiveStor Ultra to pick which Component Objects to move, it just needs to look in its local NVMe-based database.

Sonexion system’s modular, pre-integrated and compact design keeps costs low while delivering the right performance for analytics, compute clusters and supercomputers of all types. Performance and capacity scale efficiently in modular building blocks, reducing the number of hard disk drives needed to achieve sustained performance at scale. Performance is optimized end to end, from the compute clients to the network to the storage subsystem, based on the application workload.

Data protection is provided through Grid RAID, a declustered parity form of data protection that speeds up rebuild times three and a half times over traditional RAID, while maintaining Lustre performance.