Recent content by spirit

yes. (1450 should be enough with vxlan.) another way : you need to increase your physical network and hypervisor nics with minimum 1550 mtu. vxlan need 50 more bytes for tunneling

kenrel 6.8.4, ksm enabled + numa balancing=1 : no problem after 1 week, tested on 16 nodes with 1500vms

your config is ok, maybe try to change bridge-vids 2-4094 with bridge-vids 2504 Some nic model don't support too much vlan (like 256max number of differents vlan).

you can add in /etc/network/interfaces iface <vnetname> address 192.168.20.21/24

http or https ? if it's only ssh/https or other ssl protocol, it could be a mtu problem with fragmentation. (try mtu 1400 in your vm nic config for example)

This is simply the virtualisation of iops, this add latency. (Look at your vm cpu on the host, maybe you'll have 1 vmcore at 100%) do you have really have applications doing more than 16000 4k sync write with iodepth=1 ? of course, It'll increase with parrallel write. and generally...

maybe try the new 6.8 kernel (apt install proxmox-kernel-6.8). It's resolving some freeze with numa && ksm, maybe it could be related to your case too.

do you have tried through esx directly ? going through vcenter is known to be slow.

Hi, no problem since 3 days now. I'm going to upgrade all my clusters to kernel 6.8.

nftables is coming soon, and it should avoid extra fwbrX bridge, so it should work with fragmentation. But try to avoid fragmentation please ...lower your mtu ..

I have upgrade 1 server to 6.8 for testing, no problem (lag/freeze) with numa && ksm enabled since 24h ! I'll recheck monday

it should be resolved on 6.8 only.

if you only have 2 nodes, you need to use CLI to move vms. (it's not possible with gui). when a node crash: 1) verify that the node is really crashed and vms are not running anymore 2) on the other node : #pvecm expected 1 (to be able to write in /etc/pve ) 3) on the other node : #mv...

iperf3 is not multithread (even with -P option), so check that you don't have 1core at 100%. iperf2 is multithread is you use multiple stream with -P (and increase queue in vm nic options) ksoftirqd is the handling of network interrupts. a vm can do around 2millions packet per second by...

Currently, windows allocate all memory at boot. (filling all memory pages with zero). The only way to release them, is to use balloning manually. (set min memory + shared=0). Note that ksm works also fine, when you reach 80% memory usage, I'll deduplicated all zeros memory pages, but it'll...