When we wish to compare a Scale-Out NAS and especially the one developed by Rozo Systems, it’s key to understand where the product plays in term of market as it explains why some developments have been made. Key features are also fundamental in order to understand differences and limitations.

We use 6 attributes to classify and compare RozoFS with some competition offering: Performance, Scalability, Durability, Accessibility, Manageability and Cost efficiency. You will see in the table below that we mention Object Storage, this is just because, object storage solutions, at least many of them, provide erasure coding data protection mechanism. Outside of that, they can’t be considered as a competitive solution.

competitors

  • Performance is a relative perception for special use cases except when you wish to beat records. Here we consider that NAS, Scale-out NAS and RozoFS are pretty good in this category even if Scale-Out NAS and RozoFS are far better just by the capability to aggregate throughput or IOPS let’s say performance across nodes. For instance, with one server using SSD and CPU at 2.4Ghz, RozoFS is able to create 38K files per file system per second and 40K deletion per second. Wow, super fast operations. Object storage is not designed to be a fast storage but a capacity oriented one. No comparison at all in that first category.
  • Scalability is in favor of object storage except when you consider RozoFS as the product really scales. By design object storage solutions are built to store massive amount of files and huge amount of PBs. On the Scale-Out NAS side, Isilon cluster is limited to a few dozens of PBs but the majority of their deployments and use cases fit perfectly in that range. Qumulo is also good at scale. RozoFS breaks these limits and supports 1023 file systems, 255 volumes, 2ˆ59 files per file system and 16 Exabyte file system with 6TB drives. For Scale-Out NAS, the real performance attribute is the capability to continue to deliver performance incrementally with the number of nodes.
  • Durability is interesting especially with the consideration of performance. It’s not about just data protection as many products can do the job. Here we refer to efficient protection during applications or users services meaning that the failure – node, disks… - must be as much as transparent as possible. Replication is very good, very well proven and adopted, no doubt, but it works perfectly for small files. When files are big, it takes times to maintain the number of copies. The other drawback is the hardware overhead you need to consider to maintain the 3-way copies. Imagine, you have 1PB of data, you must acquire 2PB extra to reach the protection level. So at scale, there is no doubt, erasure coding techniques are the ones you need to consider, they offer superior protection with less hardware overhead. They have two drawbacks, doing erasure coding, let’s say classic ones such Reed-Solomon, consumes CPUs cycles and degrade performance and can’t be adopted for small files. Even for large, they impact the service the cluster should eviler outside. A recent discussion with an Animation Studio confirmed that they use replication whatever is the size of files is as they don’t care about cost. Isilon for instance uses a dual approach with threshold management, below this value, product uses a replication method and above this value, it does erasure coding. RozoFS uses a special erasure coding technique named Mojette Transform, patented and invented at University of Nantes and based on very simple and efficient algebra, that encodes every data submitted to the system. This very fast technique allows Rozo Systems to deploy RozoFS in very high demanding environments.
  • Accessibility is also an obvious criteria in favor of NAS as these solutions provide industry standard file sharing protocols such NFS or SMB. No need for special application integration with cost and development times associated with it, you install and use it immediately as all client machines already have built-in NFS or SMB code. Object storage must use a file storage gateway in front of them to provide this file access mechanism. No doubt, it works very well, but it adds a significant cost. Object storage vendors have made some tries but they’re all disaster as they break their scale-out approach. They have to implement single NAS head or sharding approach and it reduces drastically the scale-out affect of the solution, in fact they create their own bottleneck.
  • Manageability is a mixed property as it covers also how you integrate application but also how you deploy the product and manage it during its lifetime. NAS products are definitively better in that space, you can find some good administration tool for some object storage vendors but some others have terrible GUI and human interfaces.
  • Cost efficiency is a topic good for object storage vendor without any file access gateway thanks to the commodity servers and erasure coding techniques they use. This is also a key advantage for Scale-Out NAS solution with EC.

To conclude, it doesn’t exist so many players in the industry able to offer this kind of solution. Again RozoFS is dedicated to deliver IO Performance + Data Protection + Cost Efficiency at the same time.