SAP Hana Performance: AWS VS. GCP Showdown

By Lane Inman | Aug 01, 2019

SAP Hana Performance: AWS VS. GCP Showdown


Last month we discussed the comparison of running SAP workloads on GCP and AWS by leveraging the providers self-identified fit for SAP offerings, which revealed one of the first challenges.  A fan commented to us “Krystallize, why don’t you use GCP’s “customized” workload to match AWS…  Oh, if only it were that simple… It is not!  Interestingly enough, SAP HANA is big.  It solves so many problems with traditional databases by simply transferring the data into memory; and thus the huge size of any recommended offering by the cloud providers.

Cloud providers are rapidly approaching the billing and feature transparency of consumer cellular phone companies from the 2000s most notable for the infamous “Can you hear me now” commercials..  Platforms feature core numbers and memory counts with very little laymens terms explanations of what that means.  Much like the cell phone comparisons of the past, the standard offerings of cloud providers make this request “to match workloads” a bit more difficult.  Fortunately, GCP has introduced some flexibility by providing a means to customize workloads.

Unfortunately,  per google: Creating a VM Instance With A Custom Machine Type 

“With extended memory, you can add additional memory to a machine type with no limitations per vCPU. Regardless of the CPU platform, you can add extended memory up to a total of 455 GB per VM. For instances using the Skylake CPU Platform in zones where 96 vCPU machine types are available, you can add extended memory up to a total of 624 GB per VM instance. If you require more memory, you must use one of the mega-memory machine types, which allow you to create instances with a total of 1.4 TB per VM instance.”

So; we took to heart the “apples for apples” and ran our synthetics on the configurations listed in Table 1: Offerings evaluated.

Table 1: Offerings evaluated

Leveraging the synthetic workload created before, Krystallize Technologies applied it to the offerings evaluated in Table 1. These results are illustrated in Table 2.

Table 2: Chosen Offerings Assessed

The synthetic workload was run over the span of four hours to establish a steady compute state within the providers.  In this example, it is clear that the overall capability of the Google environment appears to be 35% greater than that of AWS for these initial runs. Workload variability was not taken into consideration, therefore further examination and publication of findings will soon follow as we determine who will take the SAP crown for performance and capability!

 

Our Performance Pros are ready to help.