A non-zero CPU ready (but host underutilized)

Marco Shaw

6/26/23, 1:34 PM

I'm having an issue with performance on a Red Hat VM, seemingly related to a Java process that seems to be quite heavy on processing and data. I'm been working through different things, but nothing has improved the performance so far.

I currently have this VM on a host, and have confirmed that I have leftover virtual processors once I calculate the core/socket/threads.

I don't understand how/why the "CPU ready" of the VM would be non-zero when the host is underutilized for vCPUs.

Won't a non-zero value potentially mean the VM is getting shifted around semi-randomly to different CPUs?

Perhaps I'm missing something obvious... I guess I'd want to try to pin the VM to same vCPUs as a test.

0 + 0

vmware-esxi

John Mahowald

6/26/23, 3:37 PM

Please describe the performance issue. How is the performance inadequate for the user? In general, what is the application doing, with how many threads? What is the CPU socket and core configuration of the host and guest? What version of guest operating system distribution (kernels and profiling tooling changed a bit over the years)?

Marco Shaw

6/26/23, 5:44 PM

An application based on Java runs 2x (or more) slower. I believe most of it is single-threaded as it's quite a complicated system. It's a newer HPE system so in the 64 vCPU range (with hyperthreading). The VM is Red Hat 7 (tried with 6,12 and even 24 vCPUs). I know I'm either not going to say enough (or say too much to describe the whole thing). My main question/point is to ask about the ESXi "CPU ready" stat. I wouldn't expect it to be non-zero when I haven't overprovisioned.

Marco Shaw

6/26/23, 5:50 PM

Can no longer edit... This all started with a P2V. We have a partner in a similar situation, and they are fine, but they use HCI (with local dedicated storage), and we do not (FC switching to a shared SAN).

John Mahowald

6/26/23, 5:51 PM

Half the speed of what? What is your baseline for comparison? And what is a typical value of CPU ready, 5% is quite a bit different than 50%.

Marco Shaw

6/26/23, 5:53 PM

I thought my question was simple... Should an underutilized ESXi still have a non-zero CPU ready? That's really all I want to know. I'll see if I need to get into the whole back story as I continue to search.

Score:0

Server

John Mahowald

6/26/23, 5:58 PM

VMware suggests a healthy CPU %READY is "under 5%". However, it can be non-zero even when CPU is not oversubscribed. It is possible for the hypervisor to not be ready on every cycle that the guest makes a request of it.

Whether this matters to performance as seen by the end user depends on an enormous number of variables, and is extremely specific to applications and workloads.

0 + 0

Elon Musk

I sit in a Tesla and translated this thread with Ai:

EN: A non-zero CPU ready (but host underutilized)

TH: CPU ที่ไม่เป็นศูนย์พร้อม (แต่โฮสต์ใช้งานน้อยเกินไป)

RO: Un procesor diferit de zero gata (dar gazdă subutilizată)

RU: Ненулевой ЦП готов (но хост недостаточно загружен)

VI: Một CPU khác không đã sẵn sàng (nhưng máy chủ được sử dụng không đúng mức)

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.