ScyllaDB Gives Cohabitation of OLAP and OLTP A Shot
ScyllaDB today announced the availability of a new feature of its Apache Cassandra-compatible database called “workload prioritization” that it says will eliminate the need for organizations to maintain separate database clusters for analytical and transactional applications.
There’s no denying that analytic (i.e. OLAP) and transactional (i.e. OLTP) workloads put different demands on the underlying database. They have different I/O patterns, different latency expectations, and different uptime requirements. Users might be interested in the same underlying data, but the characteristics of OLAP and OLTP workloads are so different that companies traditionally run them on separate hardware and then move data between them using ETL or ELT pipelines.
Of course, that doesn’t mean that some companies haven’t tried running OLAP and OLTP workloads on the same database. Plenty of companies have given the idea a gander, for better or worse, which has gone by sevearl clever names, including “translytics” or Gartner‘s HTAP, or hybrid transaction/analytical processing.
But the prospect that the production database of a large multi-national company can be brought to its knees by the careless act of a newbie analyst writing a large, malformed SQL query is typically enough to strike fear into the hearts of even the most battle-hardened DBAs, so segregation remains the order of the day.
Now ScyllaDB says it has addressed some of the concerns of workload cohabitation with its new workload prioritization offering, which effectively puts a limit on OLAP or OLTP workloads in NoSQL databases managed with its software.
The key to workload prioritization, ScyllaDB says, is the administrator’s ability to allocate “shares” of the total resources of the cluster to each workload. This allows a single Scylla database cluster to serve heterogeneous workloads while meeting performance expectations, even with traffic spikes and background maintenance operations throwing curveballs at database, the company says.
ScyllaDB co-founder and CEO Dor Laor says the new workload prioritization feature will allow companies to run OLTP and OLAP on the same cluster, thereby simplifying the overall database environment and saving money in the process.
“Database users are tired of dealing with sprawling topologies, tweaking clusters, and customizing them to support different workloads,” says ScyllaDB co-founder and CEO, Dor Laor. “With our workload prioritization feature, they don’t have to duplicate resources, isolate analytics workloads or dedicate an entire cluster to every microservice. They can consolidate their data in a single cluster, get rid of idle hardware resources and dramatically reduce maintenance costs.”
Laor and his colleague Avi Kivity, who were instrumental in the development of the KVM hypervisor, developed ScyllaDB to be a wire-compatible, drop-in replacement for Apache Cassandra, the wide column NoSQL database that has been widely adopted to serve data to numerous large-scale websites and mobile applications.
Laor told Datanami last year that, while he admired Cassandra’s overall architecture and specific capabilities – notably its scalability and data replication attributes – he found that there were some major flaws in Cassandra’s overall implementation. So he sought to do a total rewrite of ScyllaDB in C++ (compared to Cassandra’s Java), and the result is Scylla, which Laor claims holds a 10x performance advantage over Cassandra.
ScyllaDB probably won’t be in Washington D.C. next week, where much of the Cassandra world will be for DataStax Accelerate, an inaugural Cassandra event created by DataStax, which is the commercial company behind the open source Apache Cassandra database. In addition to welcoming Cassandra users like Netflix, Yelp, T-Mobile, Home Depot, and Walgreens, DataStax is expected to have some Cassandra-related news to share.