Resolve an issue with slow IO speed within KVM guest

On one of my KVM VPS hosts we have an issue where the read/write speeds on the host are much faster than the speeds achieved within the VM guest. I have set the IO policy to thread mode and the cache setting to write back, however the speed is still no where near that is achieved on the host. I'd like someone to take a look at the issue and resolve it, or at least provide an explanation for why this is and how it could be resolved. The host node is live with customers using the server, so it's a live environment so it's important that there are no interruptions or reboots to the network or server. The host is AMD EPYC CPU The guest and VM are both running CentOS 7. We use virtualizor panel to administer VMs. We have escalated this to them however they say that the node has other VMs on it so this is why the guest cannot achieve the speeds we expect (this is incorrect as shown below). Virtio drivers are enabled for the guest. The host drives are NVMe in softraid 1. There are 27 other VMs on this machine but there is no read/write bottleneck on the host and this is of course shown by the much higher read/write speeds outside of the VM versus inside the VM. This is an example of the iotop command: Total DISK READ : 78.66 K/s | Total DISK WRITE : 397.05 K/s Actual DISK READ: 78.66 K/s | Actual DISK WRITE: 338.99 K/s The VMs use qemu. Example of hdparm command on host: Timing cached reads: 15132 MB in 2.00 seconds = 7578.81 MB/sec Timing buffered disk reads: 7306 MB in 3.00 seconds = 2435.15 MB/sec Example of hdparm command on guest with IO/cache policy of threads/writeback: Timing cached reads: 12176 MB in 2.00 seconds = 6094.77 MB/sec Timing buffered disk reads: 810 MB in 3.04 seconds = 266.45 MB/sec Example of hdparm command on guest with IO/cache policy of native/none: Timing cached reads: 15952 MB in 2.00 seconds = 7987.81 MB/sec Timing buffered disk reads: 842 MB in 3.02 seconds = 279.21 MB/sec As you can see, there are very dramatic differences between the host read speed and the guest. I've tried it on both ubuntu guest and centos guest and both yeild the same result. One thing i did notice when doing some benchmarking is that the guests seem to indicate "VM-x/AMD-V : ❌ Disabled" - however when I have checked for the relevant kernel modules on the host and guest they seem to be enabled, so this is perhaps an issue with the benchmark script not interrpetting the settings correctly. Does this help identify any potential reason for the slow disk performance? It's also worth noting we have another VPS host which has an NVMe drive not in RAID and this performs exceptionally well in terms of IO performance. Therefore my logic is that the difference is something to do with a) softraid (but then how can it be fine on host but not on guest?), AMD CPU (doubtful?), virtualization not being fully enabled on the host. Please can you advise if you'd be able to review and resolve this.