Exam 3V0-21.21 All QuestionsBrowse all questions from this exam
Go to Exam
Question 32

An architect is tasked with reviewing the design of a VMware software-defined data center (SDDC) for a software development company. The platform is used to developing applications and services. It is important that the customer be able to accurately benchmark performance of developed applications.

The platform has recently commissioned new hosts to update the development cluster.

The development cluster host configuration is:

✑ 4 ESXi hosts with 2 sockets ֳ— 16 cores

✑ 512 GB RAM divided evenly between sockets

✑ There is no resource contention

The benchmarking cluster host configuration is:

✑ 8 ESXi hosts with 2 sockets ֳ— 8 cores

✑ 256 GB RAM divided evenly between sockets

✑ There is no resource contention

The customer is developing an application that includes a database virtual machine. The application developer states that the database virtual machine performs as required only when allocated 8 vCPUs 256 GB RAM. The database virtual machine performance meets the required levels when run from the development cluster. Performance benchmarking for the database virtual machine yields highly variable results when run from the benchmarking cluster. The application cannot be released without reliable performance benchmarking data.

What is a possible reason for the difference in performance test results between the development and benchmarking clusters?

    Correct Answer: A

    The highly variable performance results are likely because the database virtual machine breaches a single NUMA node boundary on the benchmarking cluster. The benchmarking cluster's hosts have 2 sockets with 8 cores and 256 GB RAM, which means the virtual machine requesting 256 GB RAM will need to access memory across two NUMA nodes, leading to memory latency and performance issues. In contrast, the development cluster's configuration with 2 sockets and 16 cores evenly divided with 512 GB RAM allows the database virtual machine to fit within a single NUMA node, avoiding these issues and resulting in stable performance.

Discussion
rajeshrubOption: A

Answer is A because benchmarking cluster will end up using both Physical VCPU. Since required vCPUs are 8 and 256GB memory hence well suited for single NUMA node on development cluster

primanturinOption: A

A https://blogs.vmware.com/performance/2017/03/virtual-machine-vcpu-and-vnuma-rightsizing-rules-of-thumb.html

VCIX_ChrisOption: A

As bright as the morning sun: A If the DB requires 256 GByte RAM, it has to use more than one NUMA domain on the benchmarking cluster hosts. Memory pages controlled by CPU B will be accessed by CPU A. That kills performance. On the development cluster the DB completely fits into one NUMA domain.

leoclyenOption: A

A. Cannot be D, because it is stated that "There is no resource contention".

CloudRRAOption: A

A seem right... As explained by rajeshrub below ;)

estornudoOption: A

Isn't it A?

FR_WolfmanOption: A

Answer A is correct. Interesting article on the subject --> https://frankdenneman.nl/2022/11/03/vsphere-8-cpu-topology-for-large-memory-footprint-vms-exceeding-numa-boundaries/

Zielony360Option: D

For me the correct answer is D, because there is contention of the memory in the host due to host's 4 GB RAM usage and VM overhead. Moreover, 8 vCPU VMs don't use NUMA by default (https://docs.vmware.com/en/VMware-vSphere/7.0/com.vmware.vsphere.resmgmt.doc/GUID-3E956FB5-8ACB-42C3-B068-664989C3FF44.html).

PSE_ITOption: A

A, NUMA node for benchmark cluster is fitting to VM DB configuration and can use different numa nodes with lower performance.

estornudoOption: D

Or maybe D because 256 GB for a VM will use the whole 256 GB host RAM ?

powered

Same feel as benchmarking wasn't 'hammered' the resource at mass level