Evaluating Performance for Cloud Data Warehousing, Part 4- Evaluate Elasticity

In my last blog post, I described methods for compiling and presenting results from your cloud data warehouse testing. I provided a framework for single-stream timing and workloads under concurrency. In this blog entry, I’ll focus exclusively on evaluating the all-important capability of most cloud data warehouse technologies: Elasticity.

Part 4 (of 5): How to evaluate elasticity, so you can make educated financial decisions

First, let me describe why elasticity is so important. Elasticity can dramatically lower total cost of ownership (TCO) by allowing you to pay for only what you use when you need it, and radically boost performance on demand by scaling the environment to meet the demand of your queries at run-time. Therefore, it is very important to evaluate this aspect of any cloud data warehouse provider - this is a clear differentiator.

There are at least three questions that should be drilled into for evaluating elasticity:

1. How are performance and throughput affected by cluster size?

For this evaluation, test a workload using different sized clusters and watch your performance and throughput as you do so. Does it improve as your expand the cluster size? Does your time-per-query decrease if you add more nodes to your cluster? Measure and document these scenarios.

2. Do throughput spikes benefit from multiple clusters?

Some cloud data warehouse providers offer the ability to spin up additional clusters if there are spikes in your workload. Test this, measure the results and document your scenarios.

3. Can the system support multiple workgroups all accessing the same data?

See if you can run multiple workloads against a single database using multiple clusters vs. a single, larger cluster. Measure and document a few scenarios related to multiple workgroups accessing the same data so you can determine whether the provider’s solution will meet your broader team needs.

Elasticity in the cloud is a major technological advancement and a critical element of today’s cloud data warehouse value proposition. A pay-only-for-what-you-need model that can expand in an instant to meet your increasing demand sounds almost too good to be true, but just about every provider says they can do it. Trust, but verify… hopefully this information gave you some tips to lead you on your way to finding your best option.

Stay tuned for the final part of the series. 

In case you missed it, check out Part - 3: Compiled Results.