virtualbox.org

Hi,

I tried to measure the throughput of a host-only network on my laptop. The laptop runs ubuntu-10.04, 32 bit. The vm runs ubuntu-10.04, 32 bit, too. I'm using VirtualBox-3.2.8.

For doing the measurement, I installed an FTP server within the VM. I created a 100MB file on the host and transferred the file multiple times to and from the VM (using FTP put and get).

I did the measurements using these types of network adapters:

• PCnet-FAST-III
• Paravirtualized network (virtio-net) (name manually translated, I'm using German VB)
• Intel PRO/1000 MT Server (82545EM)

The results are kind of strange. The "Intel PRO/1000" is clearly the fastest adapter, especially when looking at file transfers from the VM to the host (roughly 100Megs/second). The "virtio-net" is very slow when doing file transfers from the host to the VM (roughly 10 Megs/second). I'll add the exact numbers using an attachment file.

To me, these results are a surprise. I expected "virtio-net" to be the best choice.

Do you share these expectations? Do you have the same experiences? Which network adapter type do you typically use?

Thans a lot for your opinions and explanations,
Uli.

Benchmarking tools run inside a VM often gives you nonsensical results, because the virtual timers are not a reliable and true time reference.

Hi Mpack,

mpack wrote:Benchmarking tools run inside a VM often gives you nonsensical results, because the virtual timers are not a reliable and true time reference.

thanks for pointing on this. You're absolutely right. That's why I executed the FTP client on the host system. So all timing and throughput values come from the host, now from the guest.

Best regards, Uli