Aerospike Inc., which sells a commercial version of its namesake open-source, scalable, real-time NoSQL database management system, today unveiled a new version that it said is the first real-time distributed database to guarantee strict serializability of ACID transactions at a small fraction of the cost of competing systems.

Strict serializability is the strongest consistency guarantee for ACID transactions, which combine atomicity (all-or-nothing transactions), consistency, isolation and durability. It ensures that transactions appear to execute instantaneously in a total order while also respecting real-time ordering, meaning that if one transaction completes before another starts, the second transaction must see the effects of the first.

That capability is considered critical in high-volume online transaction processing scenarios. For example, if a shopper’s purchase maxes out a credit card, the credit approval application needs to be aware of that fact before authorizing a second transaction.

“Traditional databases have done this for over 30 years,” said Srini Srinivasan, Aerospike’s founder and chief technology officer. “NoSQL databases relaxed consistency for the sake of performance. Algorithms from the old days didn’t focus on performance but on getting things done right.”

Aerospike “leveraged our ability to use the latest technologies in hardware and networking and invented algorithms that enable us to provide strict serializability,” Srinivasan said. “Many are patented.”

Finding balance

Srinivasan said consistency takes a toll on performance, so NoSQL database makers have worked for years to achieve a balance. With this release, he said, “we have preserved performance while adding consistency.”

Many industries need strict serialization. Telecommunications firms, for example, typically have multiple accounts for each customer based on the services provided. “There are regulations that require these companies to change information on multiple accounts at the same time in the same database while serving customers in real time,” Srinivasan said. “We have customers who run tens of millions of transactions per second. This will let them adopt new use cases without inhibiting performance.”

Aerospike achieved consistency for single-record requests across millions of transactions per second with sub-millisecond latency in 2018. The Database 8 release expands data consistency guarantees to distributed multi-object transactions with low latency, even while serving massive concurrent connections, the company said.

“There are a number of other systems that provide strong consistency, but we expect our performance to be significantly higher,” Srinivasan said, citing MongoDB and CockroachDB as examples. He said Aerospike intends to publish performance figures that prove that assertion but isn’t doing so right now.

Performance tradeoff

He acknowledged that guaranteeing strict serializability across multiple transactions extracts a performance penalty. “Every time you change a record, you create a side note in a transaction monitor that maintains the state of the transaction, so there’s overhead,” he said. “We’ve minimized it, but our multi-record transactions will be somewhat slower than single-record.”

Aerospike simplifies OLTP development by moving the tasks of building and maintaining transaction management logic from the application to the database so developers can access the functionality via application program interfaces. Spring developers can immediately write to the same Spring Data transaction APIs they already use, and Java developers can code to the standard Spring Framework transaction management APIs without knowing anything about Aerospike internal processes.

Aerospike’s multimodel database engine supports document, key-value, graph and vector data types to allow developers to choose the best data model for each use case. The company has raised $241 million in funding, including a $109 million round last spring.

Image: SiliconANGLE/DALL-E