0

Greetings and salutations,

I would like to ask you for help with a curious situation what definitely caught my attention. We bought new hypervisor servers with two AMD EPYC 9174F processors (16 cores, 4.1 GHz, so 32 cores in total ). We are using Hyper-V. The hypervisor hosts only a single VM assigned 16 virtual cores. The VM is running Windows Server 2022 Standard and SQL Server 2019 Standard CU27.

The VM was moved to this AMD hypervisor from another hypervisor, which was equipped with two Intel Xeon Gold 6346 processors (16 cores, 3.1 GHz, 32 cores in total). When monitoring the resource usage of a VM that is now running on the new AMD hypervisor, I see very uneven CPU usage that was not the case on the previous Intel hypervisor.

As you can see in the images from Task Manager and Perfmon, the first 8 cores are significantly more loaded against the other 8. While the core for example core 0 and 1 are on average loaded at 66%, cores 14 and 15 are loaded at 2-4%. The SQL Server ERRORLOG clearly states that all 16 cores are in use, which is confirmed by querying the sys.dm_os_schedulers object.

CPU-Z's CPU benchmark is able to load all cores to 100%. However, SQL Server does not use/load some cores for some reason. SQL Server perfmon couter Batch Requests\sec is around 500. MAXDOP is 8.

Does anyone have any idea how I should go about solving this situation please?

Perfmon - low cpu core usage Perfmon - high cpu core usage Task manager CPU utilization dm_os_schedulers Core usage last hour

Improve this question
0

2 Answers 2

Reset to default
0

Both processors are frequency-optimized SKUs well-suited for SQL Server usage, but I think you are missing the Intel Speed Select Technology that is not present on that AMD processor. The Intel Speed Select Technology - Base Frequency enables to increase guaranteed base frequency on certain cores (high priority cores) in exchange for lower base frequency on remaining cores (low priority cores). That Xeon Gold 6346 improves overall performance by boosting frequency on critical cores that is why you were seeing in your monitoring system.

Even though the AMD EPYC 9174F has higher clock speed and larger cache, in your scenario or specific server workload the Intel Speed Select Technology offers better performance despite the superior raw power of the AMD Epyc processor.

On the other hand, on database servers with 32 or 16 processors I see better performance when setting max degree of parallelism (maxdop) to 8 or 4 and the Cost of Parallelism to 50. I see SQL Server doing better on queries when distributing the processing load to a maximum 8 cores.

Improve this answer
0

Almost by accident I managed to solve the problem and I'm ashamed that I didn't solve the problem sooner.

The cause of the problem was an inappropriately set Widnows Server Power Plan. As you probably know, the default setting after installation is the Balanced mode, which, however, is extremely unsuitable for performance-intensive applications, which are database systems. Out of 16 virtual cores, 8 were in "Parked" mode. Changing the configuration to "High Performance" solved the problem immediately. It's already my routine to change this configuration when setting up each database server, and I'm almost certain I've done it on this system as well.

Improve this answer

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.