Aerospike Unifies Storage Format, Delivers DBaaS
Aerospike today announced that version 7.0 of its NoSQL database will introduce a unified storage format for data stored on SSDs drives and in memory, setting the stage for multi-record transaction support in the next release. The company also rolled out its fully managed database as a service (DBaaS) offering, with availability starting on AWS followed by the other cloud providers next year.
Aerospike founder Srini Srinivasan originally developed the NoSQL database to take full advantage of the advances occurring with solid state drives (SSDs) and NVMe drives. By bypassing the operating system and the file system and going straight to the page-addressable SSDs, Aerospike was able to deliver data at near in-memory speeds but with the cost profile of SSDs.
The company also was an early supporter of Intel Optane persistent memory (PMem), and it also gave customer the option of storing all of their data in RAM for the ultimate performance (spinning disk isn’t really a thing with this speedy database). That gave the company three different storage engines and formats. In a previous release, it consolidated SSD and Optane formats, leaving two.
With today’s launch of Aerospike version 7, the company has consolidated the in-memory and SSD format as part of its unified approach to storage. “We call it the unified storage model, but it’s really the shared memory view of things,” Aerospike Chief Product Officer Lenley Hensarling says.
Having a unified storage format gives version 7 customers some useful capabilities, including the ability to perform a warm restart of in-memory data sets, which will be handy when customers need to bring their server offline, such as for an operating system upgrade or to apply patches. “You can restart these things a lot faster,” Hensarling says. “So that’s a big win.”
Another side effect of having a unified storage format is the capability to do in-memory compression of data. Users can choose from LZ4, Snappy, and ZStandard compression algorithms, which will enable them to consolidate storage engines for data in-memory and data on SSDs, thereby saving costs, the company says.
“We’re handling things like the defragmentation and the persistence of things in a common way and doing that all in-memory now,” Hensarling tells Datanami. “So then we just stream the data out when we have storage in memory, and that just speeds everything up again and makes it a lot more efficient.”
While warm re-starts and in-memory compression are nice-to-haves, the big “get” that a unified storage format brings Aerospike is the capability to introduce multi-record transactions. A unified storage format was a necessary first step for multi-record transactions, which Aerospike intends to deliver in 2024 with the 7.1 release, Hensarling says.
The addition of multi-record transactions will be a game-changer for Aerospike, the CPO says. “To have a commit-rollback kind of model, where you’re updating multiple records and get them to be strongly consistent and make sure it’s all done, and if not, roll it back–that’s something that really expands the use case coverage of our database,” Hensarling says.
Customers typically adopt Aerospike when they need very high IOPs and have hit performance walls with other databases, the former Oracle product manager says. Delivering a unified storage format that enables customers to maintain multiple database namespaces (SSD, in-memory, or hybrid) in a single database cluster gives customers the data flexibility they’re demanding, he says.
For instance, if a customer needs high performance and high IOPs on some of its data, it can carve that data out into its own in-memory namespace while leaving the rest of the data in another namespace on SSDs. But all of the namespaces are managed within the same database cluster.
“To manage that as one cluster is a big win for us,” Hensarling says. “It’s a management win not having to have separate clusters to manage. And it allows them to just tune the storage model by the work that they’re trying to do.”
The company is also rolling out its DBaaS offering. Previously, the company offered a managed database service, but customers were responsible for the underlying infrastructure. With the new Aerospike DBaaS running on AWS, customers don’t have to worry about the underlying hardware, freeing them up to focus on developing applications, Hensarling says.
“We’ve got a range of instance types and clusters. You can just pick a few things off of a menu and then, bam, you have a cluster. We run it, and you just get endpoints to the database,” he says. “If they want to move to a larger cluster size, they don’t have to go get instance types. They don’t have to do anything. They just say ‘I want to expand it’ and boom, it gets expanded.”
The Aerospike DBaaS will be based on version 6.4, so it won’t have the new unified storage format. It also won’t offer the full gamut of data models that the database offers to public cloud and on-prem customers, including key-value, document, time-series, and graph models. The DBaaS offering will only support key-value and document storage to start off, but the other data models will be added over time.