So...how did these monster servers end up giving the performance you were after in production?
Oh and I intel support basically blaming your using of their products for the shity performance or its the open source. Really now Wintel
Yeah that seemed like a bit of a throw away comment. I've hit this same issue and have read through the admin-guide and found it hard to abstract an answer from it. I ended up enabling sr-iov on my card and and bridging and NATing on different VF's.
I later found this but haven't confirmed it...
I had a win 10 vm with a GPU passed through along with a physical disk. I used sr-iov to pass a VF to the vm for north south and a Virtio Nic for east west. I also passed a dongle for keyboard and mouse. Assigned 4 cores and 8G memory. I didn't look into pinning CPUs and hugpages.
I used this...
You can't passthrough devices in the same IOMMU group to different VM's at the same time
"
/sys/kernel/iommu_groups/1/devices/0000:00:01.1
/sys/kernel/iommu_groups/1/devices/0000:02:00.1
/sys/kernel/iommu_groups/1/devices/0000:01:00.1
/sys/kernel/iommu_groups/1/devices/0000:00:01.0...
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.