Search results

do you have differents clusters by physical sites ? or is it a streched cluster with shared/replication storage ? (Because technically, pdm is able to do cross cluster migration with bridge/vlan remapping, at least in command line. (I don't remember if the option is exposed or not)

Generally, the overhead is more in pps than bandwith. (try iperf with 64bytes packets for example, you should see a bigger difference, mostly because of cpu/core limitation).

could be an mtu problem on your network, as firewall is dropping fragmented packets.

use vmbr0.110 for example, instead eno1.110. (if you are tagging the physical interface, the vlan traffic is "catched" and never reach the vmbr0, so it can't be used by the vms)

fleecing is like a cache between your vm and the backup storage. so if the backup storage is fast, the fleecing can be 0 used in size. but if the backup storage is really slow, at worse, the fleecing volume can have the same size than the vm backuped disk. (for each vm disk, a thin...

no, you never need to remove a ceph osd or mon. Just follow the upgrade doc, they are no missing step.

you need an 3rd DC for an extra ceph monitor + proxmox corosync qdevice. (but you can keep osd on 2 DC only, with ceph stretch mode + replicat4 (2 copies on each DC)

with 2 nodes, you can, but you'll need each time to unlock /etc/pve readonly manually if one of the node goes down. (#pvecm expected 1) So 3 nodes minimum to keep quorum && write to /etc/pve/ if 1 node goes down

1) on pve7, check that your ceph is 17.2 quincy. (upgrade ceph if needed) 2 ) upgrade all your your proxmox nodes to 8. (keeping the same version than ceph). 3 ) upgrade ceph to 18.2 or 19.X if you want

3/1 should be enough if you want to loose 2 nodes. (but risky if you loose a third node at the same time). 4/2 is safe

not sure it's related, but frr don't support ipv6 for underlay with evpn https://github.com/FRRouting/frr/issues/5885

disk corruption in the vm is really strange. in the worse case, the ceph should be readonly some some blocks of the vm disk.

and you cluster seem fine, no pg down, just a simple warning.

oh yes, sorry . so with 5/7 mon online, it's ok.

do you have tried to enabled fleecing option ? (it should help is your backup storage is slow)

you can do both, multiple vlans inside 1vxlan tunnel is working , but I think than multple vxlan/vnet by network is better. (not sure about nic offloading performance with vlan inside vxlan)

I don't think that search domain is already send in dhcp options. (the code is /usr/share/perl5/PVE/Network/SDN/Dhcp/Dnsmasq.pm) But it should be defined at zone level (you have a dns suffix option). they are no dhcp options configurable in subnets.cfg currently.

maybe check your cpu stats if you don't have a core at 100%, as it's quite possible that old nics don't have the RSS feature compatible with vxlan, so are not able to dispatch vxlan traffic across multiple cores.

maybe in the future, but for layer2, this will need a central gateway somewhere to do the nat. (so some kind of vm or router appliance managed by proxmox) the options are generic across differents zone type, maybe it could be possible to disable it indeed.

I really don't known how vxlan are performing with such "bit old" -> 2010 cpu ;) also modern nic have vxlan offloading, so it'll be full cpu here for vxlan encapsulation. maybe try to disable spectre,meltdown,...mitigations.