Original URL: https://www.theregister.com/2014/05/15/caringo_releases_swarm7_software_upgrade/
Caringo greases up CAStor-based software, slips out Swarm7
Same code for all nodes
Posted in Storage, 15th May 2014 16:02 GMT
Object storage startup Caringo has released its new Swarm7 storage software, promising to offer object, file, block and Hadoop support “out of the box.”
Caringo’s main technology is its CAStor object storage, which supports its own APIs and file access protocols.
Swarm7 is built on CAStor and includes the following features, according to Caringo:
- Dynamically distributed index in RAM that instantly locates and accesses objects using zero IOPS
- Architectural optimisations to increase single cluster scale to over 100PB
- Faster post-problem recovery through rapid trimming
- Faster rolling, no-downtime upgrades with a 300 per cent decrease in node boot time
- Parallel uploads for massive object sizes
Caringo says the dynamically distributed index reduces multicasting to a secondary role and CPU load by 50 per cent at nominal scale.
Three interfaces are mentioned. The first, CloudScaler, according to Caringo, is for “multi-tenant, S3-like storage, where users control security and performance behind their firewalls.”
The second interface, FileScaler, supports “SMB and NFS file protocols on top of active archives with very large hierarchical namespaces.”
A new BlockScaler interface is “for iSCSI support providing pure block over object [and will be] available in Q3.”
The actual interface protocols supported include iSCSI, HTTP, CIFS, NFS, SMB, and S3. Swarm7 also has a “read-only Hadoop file system, offering a high-bandwidth pipe straight from the storage cluster that entirely eliminates [any] separate ingest step from archive or filer into HDFS. HDFS may still be used to hold intermediate results.”
Caringo tells us Swarm is the only way to ensure uncomplicated scaling of capacity and throughput, and eliminate complexity associated with legacy storage systems as they scale. There are no single points of failure and no speciality nodes, a node being a node being a … you get the idea.
The same software appliance code runs on all nodes, and nodes can be heterogeneous in model, brand and capability. They can also be added and retired in flight.
There are also no separate (and hence vulnerable) metadata databases.
Pierre Chamberland, founder and CEO of Netmail, said: “Caringo makes it easy for us to deliver a complete email and file archival solution to our customers.”
This press release confirms the idea that, for many deployments, object storage technology swill be used underneath a file or block, or Hadoop, or HDFS, access protocol abstraction layer. Like the introduction of plastic, the new material first has to portend to be an old one, like wood, for it to be accepted. ®