Hello,
I have never tried to use SR IOV before (I have used pass-through) and I was wondering if it is a good idea, mainly in terms of reliability but also in terms of bang for the buck.
My potential use case is iSCSI between a FreeNAS VM and a Windows VM. I could just setup a virtual switch and be done with it, and with newer NICs there is allot that gets accelerated/offloaded already.
But then I thought, if I take a 10 Gig NIC and use SR IOV, both VMs could talk to each other using that instead, the newest NIC I have is an Intel 550, I also have some older 520s, MB is a SM x10srh-cf.
Then, I would be able to:
- Use the native NIC drivers
- Bypass the hypervisor, so theoretically very little overhead and really good latency
- Still use the NIC externally to the backup server (in fact, maybe better this way)
- I only have a few VMs, I might not need virtual switches at all this way
In fact, I am wondering if using a NIC this way is better than virtual switches, even if you aren't using it for anything else? (not physically connected to anything). Don't know if the power savings in the CPU emulation of switches and vNICs would offset the power used by the NIC, probably not.
Thoughts?
The reading I have done so far suggests some stability problems, but I can't tell how prevalent it is or maybe just anecdotal.
I don't want to mess with it if it is just going to be a PITA......
Thanks
-JCL
I have never tried to use SR IOV before (I have used pass-through) and I was wondering if it is a good idea, mainly in terms of reliability but also in terms of bang for the buck.
My potential use case is iSCSI between a FreeNAS VM and a Windows VM. I could just setup a virtual switch and be done with it, and with newer NICs there is allot that gets accelerated/offloaded already.
But then I thought, if I take a 10 Gig NIC and use SR IOV, both VMs could talk to each other using that instead, the newest NIC I have is an Intel 550, I also have some older 520s, MB is a SM x10srh-cf.
Then, I would be able to:
- Use the native NIC drivers
- Bypass the hypervisor, so theoretically very little overhead and really good latency
- Still use the NIC externally to the backup server (in fact, maybe better this way)
- I only have a few VMs, I might not need virtual switches at all this way
In fact, I am wondering if using a NIC this way is better than virtual switches, even if you aren't using it for anything else? (not physically connected to anything). Don't know if the power savings in the CPU emulation of switches and vNICs would offset the power used by the NIC, probably not.
Thoughts?
The reading I have done so far suggests some stability problems, but I can't tell how prevalent it is or maybe just anecdotal.
I don't want to mess with it if it is just going to be a PITA......
Thanks
-JCL
