Delivering a true Scale-Out NAS is a mission, some vendors dream about it, some others wish to be the new Isilon but they failed many times. Other chose to do object storage and tried to position and added file access mode, it has never reached a native scale-out file server behavior. But you know utopia and hopes are still two good attributes to have.

Let me summarize what I mean by Scale-Out NAS.

  1. Scale-out refers to the capability to aggregate horizontally independent nodes in a shared nothing model. Nodes are added to the pool with their own resources they control and they just shared the external network. Aggregated nodes provide more storage capacity by unification, more protection if such copy mechanism exists between nodes, more performance as potentially more nodes can deliver better access and also more compute and finally availability should be improved as well. So Scale-out has multiple advantages for sure.
  2. Providing the NAS capability while maintaining Scale-Out design and beauty is a challenge. This kind of solution must be able to expose the same file system outside the cluster via an industry file sharing protocols such NFS and multiple clients must see the same content. Any changes and updates to a file from one client must be reflected to other clients when they access the file. If such behavior is not delivered, we can’t qualify the product to be a Scale-Out NAS.

Rozo Systems chose to implement RozoFS as a Scale-Out NAS with an internal distributed design. For some of you who are familiar to pNFS you’ll recognize a pretty similar model. The model is asymmetric meaning that some machines have specific roles. It exist File Servers Heads also known as RozoFS Clients, Meta Data servers and Data servers. All these 3 roles can reside on different machines but also on same nodes. The Clients machines expose the file sharing protocol outside the cluster, so application servers talk with these machines via NFS for instance. These 3 instances can scale in different dimensions and you can add client without changing the number of data servers if you wish to keep same capacity. You can grow the capacity by adding data servers and keep same number of clients. By number you can have 256 nodes in a cluster, define 1023 file systems and 255 volumes. With very classic 6TB HDD, RozoFS can scale up to 16 Exabyte. The RozoFS model is very flexible and can be tailored to multiple use cases. I invite you to try it, contact us at info@rozosystems.com.

architecture