NoSQL Databases Resources - Page 3

Hi All! For teams running business-critical workloads, vendor support and SLA guarantees can matter as much as raw features. I've been researching the Wide Column Database category for where reviewers actually credit strong SLAs and responsive support, which tends to favor the managed cloud services here.

• Microsoft Azure Cosmos DB: Reviewers highlight its SLA-backed guarantees around availability and latency as a key reason to trust it for critical workloads, though a few note Microsoft support responses can be inconsistent.
• Amazon Keyspaces: Backed by an availability SLA with multi-AZ replication and AWS handling patching and scaling, which reviewers value for offloading operational risk. The trade-off they cite is cost.
• ScyllaDB: Reviewers point to strong community and vendor support that reduces downtime risk, plus managed-service options. The gap they flag is monitoring integration maturity versus the big cloud providers.
• HBase: As an Apache project, its "support" is community and ecosystem rather than a vendor SLA, which reviewers say is improving but is thinner than the managed clouds; worth weighing if guaranteed SLAs are the requirement.

For those with SLA-sensitive workloads, did a managed provider's SLA actually hold up during an incident, and how responsive was support when you genuinely needed it? Stories from real outages welcome.

Hi G2 community. I'm looking into wide column databases for time-series workloads, and rather than crown a "fastest" one (reviews don't carry head-to-head benchmarks), I'll go by which platforms reviewers actually run for high-write, time-ordered data with low latency, within the Wide Column Database category.

• ScyllaDB: Reviewers specifically name time-series and event-ingestion among its sweet spots, citing predictable low latency and high write throughput at scale on fewer nodes. Data-model design (partition and clustering keys) is what makes or breaks time-series performance here.
• Cassandra: Its exceptionally fast writes and append-friendly model make it a long-standing choice for time-ordered data, with reviewers praising write throughput. The trade-offs are read performance and careful key design.
• HBase: Reviewers use it for real-time read/write on very large datasets and near-real-time analytics, which fits time-series at volume; just note that aggregation/OLAP queries are slower, and it leans on the Hadoop stack.
• Microsoft Azure Cosmos DB: Single-digit-ms latency and instant scaling make it viable for time-series if you're on Azure, with the usual caveat that RU/s cost needs planning for high-ingest workloads.

For time-series specifically, what mattered more in practice: raw write speed, or how well the partition/clustering design held up as data grew? Curious which of these stayed fast once the series got long.

After analysis of the Wide Column Database category on G2, I’ve learnt that pricing model matters as much as sticker price with these databases, since self-managed, serverless, and provisioned-throughput approaches each fail differently on cost. I’m looking into how reviewers describe the actual pricing models and what they do to keep costs down, here are some great tools I found:

• Microsoft Azure Cosmos DB: Priced on request units (RU/s); reviewers stress that careful throughput planning and partition-key design are the main cost-optimization levers, since poorly planned RU/s is where the bill balloons.
• Amazon Keyspaces: Offers on-demand and provisioned capacity modes, and reviewers optimize by matching mode to workload predictability. Auto-scaling helps, but several note that the service runs costly if you don't tune it.
• ScyllaDB: The cost lever reviewers cite is node reduction; its efficient hardware use means smaller clusters for the same load, and auto-tuning cuts DBA time. Available self-managed or as a managed cloud service.
• Cassandra: Free and open source, so the cost is operational rather than licensing; reviewers optimize by right-sizing nodes, though they warn that the DevOps overhead and RAM needs are the real expense.

When you evaluated these, did a consumption model (RU/s, serverless) or a self-managed cluster end up cheaper at your scale? And what single cost-optimization change saved you the most?