Roughly six months after transitioning its Astra DB managed Cassandra cloud database service, DataStax is taking the logical next step in extending serverless support to go cross-region. Cassandra is a logical target for such a capability since it has traditionally been designed for extreme, scale-out, distributed deployment use cases requiring fast writes.
The going notion is that serverless simplifies and, for highly variable traffic loads, can reduce the cost of running databases. With globally distributed databases, the costs of overprovisioning quickly add up.
Astra DB joins a very select group of databases that are serverless and can span more than one region to varying extents. It’s currently the only one running on multiple clouds that’s in GA. Others include PlanetScale, which we covered just over a year ago that allows you to selectively shard MySQL across one or more regions; but it is still currently in beta. There’s DynamoDB, which offers a Global Tables feature for designating some tables for replication to other regions, and Cosmos DB also offers a serverless option — but both of those only operate on their respective clouds. However, other multiregion database cloud services, such as Google Cloud Spanner and Cockroach Cloud are not serverless. Serverless platforms extending across multiple regions are still fairly rare because designing them requires some fancy footwork when it comes to coordinating resources.
The new multiregion capability runs Astra DB as a single logical instance across all regions chosen by the customer; within each region, there are at least two data centers or availability zones for failover. Within the group of regions, admins configure how and where writes are committed and replicated, for failover and local accessibility. The default practice for failover is with another data center in the same region. As to replicating writes, admins can be selective so as to keep a lid on costs of write request units charged by cloud providers.
Use cases for globally distributed databases include applications requiring the ability to conduct fast writes and reads across multiple regions, without the latency associated with remote access. Additionally, with data sovereignty or residency laws rapidly emerging, having cross-region capability provides an alternative to maintaining multiple separate instances; in these cases, admins can essentially shard the database to confine subsets of data to their home regions.
Prior to the new release, some DataStax customers were already working with multiregion deployments involving cross-region replication. Infosys, for instance, set up VMs in each region where Cassandra was hosted for handling the replication for a global client. They faced headaches having maintain and keep current all the VMs and associated YAML files. With Astra DB now adding multiregion support for serverless, those complexities are buried under the hood.
At this point, the use cases for Astra DB across multiple regions is for scenarios requiring fast local access that are also sensitive to data egress costs. While Astra DB is available on all three major public clouds, there is little or no demand for so-called “inter-cloud” (running the same database across more than one cloud provider).
Disclosure: DataStax is a dbInsight client.
- cloud services
- data center
- data centers
- remote access