Which GCP instances are best for OLTP workloads?

Which GCP instances are best for OLTP workloads?
[ Guides ]

Which is best?

76 pages of extensive research into the best OLTP instance types on GCP, as well as the other public clouds.

Get your copy

Choosing the right instance type for your workload can be a tricky proposition. It’s not always clear how a particular configuration is going to perform for your workload, and running the tests to find out is time consuming and expensive.

So here’s the good news: we did the tests, and in this article, we’re going to share the results.

In the 2022 Cloud Report, we looked at the OLTP performance of 18 different GCP instance types and 36 total configurations across more than 3,000 total runs. Specifically, we tested OLTP performance and price-for-performance using a metric called Cockroach Labs TPC-C Derivative Nowait. We tested both small (8 vCPU) and large (~32 vCPU) instance types both separately and comparatively on a per-vCPU performance basis.

In this article, we’ll take a closer look at some of the most important results for GCP customers with OLTP workloads – how each instance type fared in our OLTP tests.

(The full report, which is free, contains additional OLTP results, as well as other benchmarks for CPU, storage, and networking for these GCP instances as well as dozens of comparable instances on AWS and Azure).

Overall median TPM

In terms of overall performance, the larger 32 vCPU nodes unsurprisingly outperformed the smaller nodes. For both large and small instance types, though, the overall category winner in terms of median TPM was the t2d instance, which uses AMD Milan (Epyc Gen 3) processors.

overall median tpm, gcp

The t2d instances were consistent high performers in our testing, and we saw instances running AMD’s Milan processors topping the results across all three clouds.

Median TPM per vCPU

When breaking down the performance on a per-vCPU basis, we found that the best-performing small nodes actually outperformed the larger nodes. The top four finishers all use latest-gen processors (the t2d instances use AMD Milan, the n2-icelake instances use Intel Ice Lake).

tpm per vcpu, gcp

Median TPM per vCPU with storage

We tested each instance type with both general purpose (pd-ssd) and high-performance (pd-extreme) storage. However, we found that for the OLTP workloads we tested, the high-end storage didn’t confer much of a performance benefit.

Median tpm per vcpu with storage, gcp


When looking at price-for-performance (measured in $/TPM, see the full report for details), we found that once again the t2d instances with AMD Milan processors led the pack for both large and small instances, followed closely by the n2-icelake instances.

gcp instance type price for performance, small instances

gcp instance type price for performance, large instances

Want to go even deeper?

The 2022 Cloud Report and goes into more detail on how these GCP instances performed in OLTP testing, and also includes benchmarks for CPU, networking, and storage performance. It also includes results for these same tests on AWS and Azure.

It’s also free, so check out the full report for all of the information you need to pick the best cloud and instance type for your OLTP workloads.

About the author

Charlie Custer github link linkedin link

Charlie is a former teacher, tech journalist, and filmmaker who’s now combined those three professions into writing and making videos about databases and application development (and occasionally messing with NLP and Python to create weird things in his spare time).

Keep Reading

AMD vs. Intel and more: What’s new in the 2022 Cloud Report

The 2022 Cloud Report is finally here, and we think it was worth the wait!

In this completely free 70+ report, you’ll …

Read more
How to benchmark and tune Google Cloud

Engineers on cloud performance teams can spend their entire workday tuning and optimizing cloud configurations. We …

Read more
GCP outpaces Azure, AWS in the 2021 Cloud Report

The 2021 Cloud Report stands on benchmarks. Now in its third year, our report is more precise than ever, capturing an …

Read more