Caringo added support for iSCSI and Hadoop and took steps to increase performance and scalability of the software formerly known as Caringo CAStor.
"Our name was getting old," said Adrian Herrera, Caringo's senior director of marketing. "It was time to freshen up our look."
Caringo has added dynamic distributed indexing in RAM to Swarm, making the software faster and giving it the ability to support more than 100 petabytes (PB) vs. the previous limit of 60 PB. The distributed index helps to instantly access objects without using any IOPS. Herrera said the software previously used multicasting to locate data, a process in which a broad request for data was sent to every node. Now the distributed index sends out a unicast request because the software can sense where the data is and can go directly to the node with the data. The company claims this approach reduces CPU usage by approximately 50%.
"The problem with multicasting is [that] when you get into large clusters, it slows down the system. So we got rid of it," Herrera said.
Caringo also enhanced Hadoop support in Swarm, adding the ability to ingest data directly from Hadoop Distributed File System into the storage cluster. Swarm now does parallel uploads, breaking up large objects in the terabyte range and streaming them into multiple nodes simultaneously.
Swarm's CloudScaler cloud gateway now has full support for Amazon S3 and within a month it will include a BlockScaler to support iSCSI, Herrera said. The Caringo File Server has been renamed FileScaler.
Marc Staimer, president of Dragon Slayer Consulting, said the name change was smart. Caringo's CAStor was named for content-addressed storage (CAS), an archiving technology no longer in vogue. "The product was originally in CAS and that market is a non-event," he said. "It's more of a feature, so changing the name made sense."
As far as the product enhancements, Staimer said, "They made it more intelligent in a way that it is more elastic. It has the ability to treat a lot of resources at once. Not everything is treated in the same way. You use different resources based on the workloads. It's policy-driven. The product is more flexible."