0

I have a Hyper-V host on win2012R2 (Version 6.3.9600) with two gigabit network adapters, one of them connecting to a gigabit router (the another one is currently out of use because lack of a gigabit switch near to the server yet).

I'm running a CentOS 7 guest to run netatalk as a TimeMachine service for my Mac.

I use the network adapter on the guest as an external bridge to my router (normal virtual network adapter, not legacy (!)).

My problem is if I running a network bandwidth test between the host and the guest i just get 10MB/s (if I run python's speedtest-cli on the guest I got 153MB/s and on the host 240MB/s regarding to my broadband).

The commands I use: Hyper-V host:

nc -v -l 2222 > NUL

And the CentOS7 guest:

dd if=/dev/zero bs=1024k count=1024 | nc -v 192.168.250.195 2222

If I set up an internal virtual network adapter between the host and the guest I get 130MB/s-300MB/s with the same commands.

The guest has dynamic memory starting at 512MB, and 2 of the 4 CPU cores, but I don't thinks this matters in this case.

dd if=/dev/zero bs=1024k count=1024 | md5sum

Gives 85MB/s-120MB/s

And a normal

dd if=/dev/zero of=test bs=1024k count=1024

Gives 350-450MB/s (Dynamic VHDX on a SATA3 disk)

How could it be? And what can I do for it? I've already ordered a gigabit switch to use the second physical network interface only for the guest but I'm afraid that this isn't the main cause of this error.

Unfortunately I not have been able to run dd over netcat from my mac to the guest because I haven't been at home for couple of days.

Thank you for your answers.

Ewert
  • 5
  • 5
  • Hi Ewert, what is the vendor and model of your Ethernet NIC? – Dylan Knoll Jul 20 '15 at 21:50
  • Hi, they are Realtek® 8111G – Ewert Jul 20 '15 at 22:22
  • Are Virtual Machine Queues enabled on the physical adapter in host network properties? This must be supported by the NIC and it appears it is not in the case of the RTL8111G, could be causing performance problems. – Dylan Knoll Jul 20 '15 at 22:30
  • Yeah that can be a problem, but VMQ is unneccesary for this host, because only this CentOS guest needs much traffic. The couple other guest are used for monitoring via snmp. – Ewert Jul 21 '15 at 14:34
  • I'll give a try to the another NIC in SR-IOV mode without host sharing, gaved only to the CentOs guest as soon as i'm phisically near to the server and when I got the new switch. Or maybe I'll setup NIC teaming between the to apadters. – Ewert Jul 21 '15 at 14:37
  • Well, I finally got the new switch, but that is not what solved the problem. I've found out, that I've given static IP for the CentOS guest, as usual, but when i've tried a same instance of this guest with DHCP networking, and ran dd, it was be able to use the gigabit network (113MB/s). When I switched to static ip, I got 100Mbits again. It's really strange, but it is working now properly. – Ewert Jul 23 '15 at 13:28

1 Answers1

0

I don't know why, but with DHCP networking it's running w/ gigabit speed. With static IP it got 100Mbits again.

But it's working.

Ewert
  • 5
  • 5