Join Log in
Score:0
avatar Server

KVM guest with multiple vcpu allocated produce poorer results

cn flag
user161458

My KVM host have a 4 cores cpu (a bit weak) and run a kernel configured for real time (with preempt_rt_full, isolcpus, no_hz full). The cores 2 and 3 are isolated.

My guests kernel are also configured for real time (with preempt_rt_full and no_hz_full).

When running two guests with 1 vcpu (one guest on each isolated core) the latency is not noticeable. Here are the result of cyclictest:

VM1:/home/debian# cyclictest --default-system
T: 0 ( 737) P: 0 I:1000 C:   50458 Min:      17 Act:  207 Avg:  187 Max:    1275

VM2:/home/debian# cyclictest --default-system
T: 0 ( 737) P: 0 I:1000 C:   58514 Min:      15 Act:  193 Avg:  181 Max:    1186

When running one guest with 2 vcpu (on both isolated core) the latency is really high. Here is the result of cyclictest:

VM1:/home/debian# cyclictest --default-system
T: 0 ( 737) P: 0 I:1000 C:   40795 Min:      11 Act:  207 Avg:  697 Max:    775899

I tried unsuccessfully multiple solutions:

  • Running the 2 vcpu configured guest without the cpu affinity (so on core 0 and 1)
  • Running kvm daemon (libvirtd) on the same cores as the guest (so on core 2 and 3)
  • Removing the unnecessary controller (tablet, ...)
  • Increasing the RAM allocated to the guest

Do you know where I failed ?

12
0 + 0
ธงชาติไทย
Elon Musk
I sit in a Tesla and translated this thread with Ai:

mangohost

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.