ARA-C01 Question 345
Select 2Impact of costsA global retail company runs hundreds of ELT jobs in Snowflake every hour. The architecture team notices that compute spend has increased sharply over the last month, even though total data volume has remained nearly constant. Investigation shows many pipelines are configured to use large virtual warehouses, several jobs run for less than 30 seconds, and dashboards query the same transformed tables repeatedly throughout the day. The company wants to reduce Snowflake cost without materially affecting service levels. Which TWO actions would most directly reduce ongoing cost in this scenario?
- A
Consolidate very short, frequent jobs where practical to reduce the impact of warehouse resume activity and per-second billing after the 60-second minimum
- B
Replace all dashboard workloads with larger multi-cluster warehouses so queries complete faster and therefore always cost less
- C
Use appropriately sized warehouses for ELT instead of defaulting to large sizes for every pipeline
- D
Disable result caching so warehouse resources are used consistently and performance is easier to predict
- E
Create additional copies of transformed tables for each dashboard team so they do not contend for the same micro-partitions
Show answer and explanation
Correct answers: A, C
Explanation
The strongest cost optimizations here focus on compute efficiency. Snowflake credit consumption is driven heavily by warehouse size, runtime, and resume patterns. For short-lived workloads, the 60-second minimum charge on warehouse resume can make many tiny jobs disproportionately expensive, so batching or consolidating them can reduce waste. Separately, right-sizing warehouses is a standard best practice: choose the smallest warehouse that meets performance and SLA requirements rather than defaulting to larger sizes.
The repeated dashboard access pattern also points to the value of Snowflake caching behavior, especially persisted query results where applicable, so disabling cache would be counterproductive. Multi-cluster warehouses are useful for concurrency scaling but are not automatically a cost-saving tool. Likewise, duplicating tables generally increases storage and management overhead without addressing the root cause of compute spend.
These recommendations align with Snowflake best practices around cost governance, warehouse sizing, auto-suspend/auto-resume behavior, and query/result reuse as described in Snowflake documentation on understanding compute costs, warehouse billing, and performance optimization.
- A. Correct.
Correct. Snowflake warehouses are billed per second, but each resume has a 60-second minimum charge. If many jobs run for only a few seconds and repeatedly resume warehouses, the minimum charge can become a meaningful cost driver. Consolidating short tasks, batching work, or reducing excessive suspend/resume cycles can lower compute cost while maintaining acceptable latency.
- B. Incorrect.
Incorrect. Larger or multi-cluster warehouses do not inherently lower cost. They may improve concurrency or reduce query runtime, but they also consume more credits. Multi-cluster warehouses are intended for handling concurrent workloads, not as a general cost-optimization mechanism. In many cases, using larger compute for dashboards increases cost unless justified by strict performance requirements.
- C. Correct.
Correct. Right-sizing warehouses is a core cost-control practice in Snowflake. Using large warehouses for all ELT jobs often wastes credits when workloads do not need that level of compute. Matching warehouse size to workload characteristics can significantly reduce spend without materially affecting SLAs.
- D. Incorrect.
Incorrect. Disabling result caching would usually increase cost, not reduce it. Snowflake's persisted query results can allow repeated identical queries to return without additional warehouse compute under the right conditions. Since the scenario mentions dashboards repeatedly querying the same transformed tables, preserving caching benefits is aligned with cost reduction.
- E. Incorrect.
Incorrect. Creating extra copies of transformed tables adds storage cost and operational complexity, and it does not directly address the primary compute cost drivers in the scenario. Snowflake handles concurrent access to shared tables well, and duplicating tables for each dashboard team is generally not a recommended cost optimization.