Storage That Scales
New SONAS release offers enhanced performance
You can scale out or scale up. Scaling up usually means a forklift upgrade to a larger, single machine. Many other industries like high-performance computing and Internet-centric companies have grappled with outgrowing the single largest machine they could buy, and instead moved to a different model. “Instead of a herd of elephants, they moved to an army of ants,” Neville says.
SONAS storage devices are designed to scale out, which means you can add (or decommission) nodes in one large, clustered system, nondisruptively. Instead of many individual machines, customers see one shared namespace with all nodes able to access all data all the time, which also results in extremely high availability. Other NAS solutions often bundle the server and storage together. The drawback is if you require more I/O capacity (server power), you must buy another, larger server and migrate everything, even if you don’t need more capacity.
“Many customers dislike the traditional monolithic NAS design,” Neville says. “They want to add storage without adding throughput or vice versa. SONAS is designed differently in that we’ve separated the storage (TB/PB) and the servers (interfaces, I/O and bandwidth) into modular pieces, which can be upgraded independently.”
SONAS has storage pods and interface nodes. Designs can be archive centric (many storage pods, few interface nodes), performance centric or both. You can transparently add more storage and/or CPU power when necessary. “This may seem intuitive in other industries, but in the NAS world, the ‘army of ants’ design is revolutionary and historically it’s been challenging to preserve performance in smaller block random I/O across a single large namespace,” he says.
The Truth About ILM
Neville says the term “information lifecycle management” is often misused in the storage industry. Other NAS offerings may tout ILM and HSM as features, but SONAS is one of the only NAS offerings in the marketplace that provides true scalability, he says. When a move process is performed on a file, you must first “find” it within your millions or billions of files. This may take days, weeks or months with traditional technologies. Neville says other NAS products often use designs that interrogate individual files to find the one chosen for the move process. “If you have hundreds of thousands of files you can do the find process in a reasonable timeframe; it may take a day,” Neville says. “If you have billions of files, this becomes ridiculous.” SONAS has a fast policy engine that can scan billions of files in minutes, and then perform many actions to them including physically moving/tiering data (without namespace changes or stubbing), deleting, replicating, etc.
Neville tells the story of a test for a customer who ran a traditional HSM move process on a large dataset and let it run for several weeks. The query never came back with a result. They then ran the same process with the SONAS policy engine and received a result in five minutes. “It blew their socks off,” he says, “it opens up entirely new use cases for managing your data. Everyone says, ‘We do tiering, HSM and ILM,’ but design matters—IBM does it differently.”
Another way SONAS ILM differs from traditional NAS offerings is by supporting tape (more than 900 different hardware devices from numerous vendors); data on tape can be stored at one-fifth the cost of disk, can be 290 times less expensive in terms of energy and can last 30 years or more. File migration to and from tape is fully automated/policy driven and results in a lower long-term TCO both in terms of media and administration. “We’re the only solution in the market that offers true, scalable file lifecycle management fully integrated with efficiency around leveraging tape,” Eceral says.