Status of network performance on OI

[dri]

Hello,
I recently discovered that the network performance in guests has dropped through the floor using VB 4.2.6 on OpenIndiana 151a7.

I'm using crossbow to map VLAN's into the guests but I also tried bridging to a physical port, still the same performance. The guest is a Ubuntu 12.04, using both Intel and virtio vnic's. Poor performance in this context is below 10MB/s using a NFS client inside the guest and around 15MB/s using netcat between hosts.

I've been running VB on OpenSolaris and now OpenIndiana for several years but it's not until now I've actually needed to do stuff outside the VM that requires more bandwidth. Do I need to start looking into using KVM or is this being sorted?

Can someone confirm or dispute the performance I'm seeing?

Thanks!

[martyscholes]

You have me curious. I run Solaris 11.1 with several guests at any moment in time, but none that are particularly bandwidth-sensitive, so I have never done any testing.

Recently we started switching to letting VB allocate the Crossbow VNICs. You say you are using Crossbow to map VLANs to the guests. What does that mean? Does that mean you allocate a VNIC in Solaris and bridge the guest to that?

[dri]

You have me curious. I run Solaris 11.1 with several guests at any moment in time, but none that are particularly bandwidth-sensitive, so I have never done any testing.

Can you please test what bandwidth you can get out of a Solaris 11.1 vbox Linux guest? Something simple, like netcat. I'm considering to switch to Solaris if I find KVM too much of a hassle to bother with.

Recently we started switching to letting VB allocate the Crossbow VNICs. You say you are using Crossbow to map VLANs to the guests. What does that mean? Does that mean you allocate a VNIC in Solaris and bridge the guest to that?

I understand that there's some voodoo crossbow driver that's included in recent vbox releases that creates crossbow interfaces automatically. I've done them manually since OpenSolaris was around by simply creating a vnic (dladm create-vnic) with the vbox generated mac-address and then bridge to that interface in vbox. I then have full control over where that interface would map to VLAN-wise.