Do I need a single Virtual Switch per VM Guest

[Berd]

Hello Peers,

I am relatively new to HPVM's We are experiencing poor network performance from VM guests at certain peak times. Both VM Host and Guest are running 11.31.

There is only a single virtual switch which is used by all guests on this host.

If I migrate a VM guest to a less busy host, the network performance improves.

Should I have more virtual switches to cope with the load ? And if so, are they ok to share the same physical NIC's ?

Thanks in anticipation,

Berd

[Berd]

Another way to ask the question, is there a optimum ratio of guests per vSwitch ?

[Dave Olker · Accepted solution]

---

@Berd wrote:

Another way to ask the question, is there a optimum ratio of guests per vSwitch ?

---

There's no documented ratio that I've ever seen because not all vSwitches are created equally.  You can build a vSwitch on a 1GB link, 10GB, FlexNIC ranging anywhere from 1-10GB, an APA trunk that could include multiple 10GB links, etc.  If you're finding a VM guest is not getting the bandwidth they need from a shared vSwitch you can take several approaches:

1)  Ensure the underlying NIC(s) backing the vSwitch are configured optimally

2)  Make sure you have the latest AVIO and networking drivers installed in the host and guest

3)  Increase the bandwidth of the vSwitch by using a faster/larger pipe to back the vSwitch

4)  Spread the VM guests across multiple vSwitches

5)  Use Direct I/O (available with Integrity VM & vPars v6.1) and bypass the vSwitch completely

Regards,

Dave