FlashBlade was built to scale to 10s of PBs of data and billions of files/objects, handling both IO and metadata scale seamlessly as it grows. An elastic scale-out design at every level, FlashBlade is simple to manage at any size. Need anything more? Just add a blade.
When data gets big, both performance and affordability matter. FlashBlade was designed around raw flash, a 100%-Ethernet architecture, inline data reduction software, and System-on-a-Chip processors to deliver both breakthrough performance and economics – you won't have to choose.
With an N+2 design for redundancy, FlashBlade can tolerate two full blade failures and many more flash chip failures without data service interruption. The latest LDPC error recovery protects against flash bit errors, and all data is encrypted all the time. Blade loss results in a predictable 1/N aggregate performance loss, and the system auto-rebuilds to heal around failures.
The future is multi-protocol, and FlashBlade was meant to speak and translate between multiple languages. Unlike legacy products where protocols are stacked upon one another, Elasticity implements a common Object Core, and builds all protocols on top as peers. This means you can read and write simultaneously in any protocol, and share data easily. It also means that future protocols can easily be added in the future.
Elastic scale-out storage services at every layer.
The scale-out object core of the system, the Elastic Core implements the basic CRUD (Create, Read, Update, Delete) semantics that all protocols are built-on. It also implements all system and flash resiliency services, including data reduction, encryption, as well as full reliability protection against flash and component failure. Future data services will also include snapshots and replication.
The extensible metadata engine of the system, the Elastic Map delivers variable-block metadata across all layers of the system. Metadata performance is treated just as critically as I/O performance, and the Map can be queried independently by the user for rich file usage and auditing analytics.
Multiple protocols are implemented as peers on top of the Elastic Core, allowing the system to present data via file, object, or higher-level analytic protocols simultaneously. Each protocol is implemented from scratch in a true scale-out fashion, and extensibility allows future protocols to be added easily.
A built-in, software-defined fabric that scales to 100s of Blades and 1000s of clients.
From client to Blade, a commodity 40 Gb/s Ethernet network ensures the fastest, cheapest connectivity that you never have to manage.
No TCP/IP or SCSI spoken here. Proprietary object messaging and distributed transaction protocols ensure fast and reliable communication between Elasticity instances running on each blade.
Separate virtual networks isolate client, management/metadata, and data transport with defined SLAs.
All FlashBlade specifications and features are preliminary and subject to change prior to the GA version release.