Technology Overview
Accessing files from a bank of computers networked together has evolved over time. It started with the Network File System protocol in the early 1980s. Now Scale-out NAS is the current approach with cloud computing and cloud storage looking to be the next milestone marker.
The main attraction of NAS has always been simplicity. Because a NAS appliance integrates the file system, server, and storage into one device, ease of deployment and ongoing administration are the attractive selling points.
However, most NAS systems are like buckets: fill it up, buy another bucket. The first bucket is great—the tenth bucket is painful. Add another bucket and spend the afternoon ladling data from overflowing buckets to the new one. Administrators spend time managing buckets instead of IT operations.
With IBRIX, the bucket simply gets bigger. It is one big bucket that uses a single global namespace, which means that adding capacity doesn't mean additional work. IBRIX Fusion automatically manages it. Google™ and Amazon™ have proven that cluster-based file storage can support massive capacities and Internet-scale performance in 7x24 operations, but their systems are proprietary. Only IBRIX offers enterprise-quality Internet-scale storage that you can buy today.
The IBRIX Approach
A new breed of NAS called Scale-out NAS has emerged to break through these scalability limits. The IBRIX approach is more scalable because it is unique.
Distributed File Metadata Access
Most scalable clustered storage uses a dedicated metadata server to keep track of file locations, unused capacity and write permissions. Centralized metadata servers are easy to build and there are a variety of techniques for mitigating server failure. Yet dedicated metadata servers can limit performance by choking when many files are created or requested or when the access pattern consists of too many small files.
IBRIX’s unique architecture, called a segmented file system (U.S. Patent 6,782,389), overcomes these problems without compromising scalability. Each storage node has its own local file system. With IBRIX Fusion, metadata operations are distributed across all storage servers. Each storage server handles its own traffic so as servers are added, performance increases.
The IBRIX “Area Code” Solution
Distributed metadata creates other issues like ensuring applications get the most recent version of a file. If all storage servers can access all files—whether local or on another storage server—all cached metadata must be updated quickly. Since file access is locked until all nodes are updated, performance suffers for popular files.
One common solution to minimize update time is to use a low-latency, backend network to update all the storage servers in a cluster. This works well for small clusters up to 32 nodes, though it’s a costly solution and adds management overhead.
IBRIX discovered and patented another solution to the problem of updating metadata operations. IBRIX Fusion separates file location from metadata currency, which is akin to how a telephone area code works by routing a call to the correct switch. When a storage server receives a request, the server checks the area code of the file requested to see if it is local. If it isn't, the server consults a map—an area code lookup—to locate the server managing that area code and forwards the request there. All files are accessed with a maximum of one hop independent of the number of servers or files in the system, reducing latency and network congestion and guaranteeing scalability.
In the early stages of our film, it was taking 10 times longer to render each frame because the NFS caches could not go fast enough to serve up the data needed. Having IBRIX in our network allowed us to get the film done on time—extremely important when you consider the billions in revenue that’s at stake from box office receipts and consumer spending on film-related merchandise. 
Vice President of Technology
Pixar Animation Studios
The IBRIX Fusion architecture doesn't need to update all metadata on all servers to maintain file system consistency. Only those servers with a copy of the file need to be updated to maintain data consistency. And if a file location changes due to a storage server failure, for example, only the "area code" map needs to be updated.
With IBRIX Fusion, all nodes work in parallel to complete tasks faster. All operations are done on a per node basis and in parallel across all storage so that the time for any operaiton does not grow linearly with the size of the file system—a BIG breakthrough.
Overall, IBRIX Fusion brings several advantages to large-scale file systems and storage. By maximizing parallelism, the architecture is uniquely scalable in performance, availability and capacity. As a pure software solution, it enables customers to choose the best hardware for their application. And with support from leading partners Dell, EMC and HP, it is a reliable solution for critical applications.