Dedicating physical NIC ports to each VE without bridging, etc. [message #15960] |
Fri, 17 August 2007 02:06 ![Go to next message Go to next message](/theme/ovz3/images/down.png) |
dranch
Messages: 33 Registered: August 2007
|
Member |
|
|
Hello Everyone,
I'm looking to see if there is any way to physically dedicate
a physical NIC port to each VE. If I can do this, it should
remove all the various complexities of troubleshooting
bridging, etc. The main reason for this is because I'm worried
about all the issues I might see when my users are messing with various things in the VEs such as altering MTUs (smaller/jumbo), VLAN tags, etc.
Xen can do dedicated NIC mode this via a hacky way of
excluding the PCI ID's from the Dom0 kernel and allowing the DomU kernels to discover then on their own. But.. I don't want to
run Xen. ![Smile](images/smiley_icons/icon_smile.gif)
For a specific example, say I have the following.
A physical machine with four ethernet ports (eth0-3) and I want:
physical port Mapped location
eth0 base machine for management and access
to all other VEs for management only
eth1 VE 100 - eth0
eth2 VE 101 - eth0
eth3 VE 102 - eth0
According to the following URL posted on the OpenVZ site, page 68: This *is* possible:
http://download.openvz.org/doc/OpenVZ-Users-Guide.pdf
"Still, there is a possibility for a VPS to directly access
a physical network adapter (for example, eth1). In this
case the adapter becomes inaccessible to the Hardware
Node itself."
vzctl set 100 --netdev_add eth1 --save
When I do this, it's doing two things:
1. For VE#100, it's actually mapping this interface in
VE#100 as eth1 when I really want it mapped as eth0.
This could give be an eth0 in every VE. The manpage
for vzctl says that all other options for netdev_add
are optional but whenever I give the full syntax,
it complains about the VE's MAC address being an
incorrect legth. So I altered the resulting
/etc/vz/conf/100.conf file as follows and
that help map things to eth0 in VE#100 but it's not
doing what I want (see item #2):
NETIF=" ifname=eth0,mac=00:18:51:7C:1C:A6,host_ifname=veth100.1,host _mac=00:18:51
:1F:F4:0B"
2. When I "ifconfig eth1 up" the interface in VE#0 (master
OS) with no IP assigned, then bring eth0 up in VE#100
with say "ifconfig eth0 1.1.0.1 netmask 255.255.255.0",
I don't seem to be able to transmit any data (no blinking
LEDs). If I run tcpdump in VE#0 on interface
"veth100.1", I *DO* see the ARP requests for my pings,
etc.
Item #2 above gives me the impression that I *MUST* configure
bridging and map this veth100.1 interface and the eth1 interface into a unique bridge group within VE#0.
Is there any way to avoid this?
--David
|
|
|
|
Re: Dedicating physical NIC ports to each VE without bridging, etc. [message #15993 is a reply to message #15962] |
Fri, 17 August 2007 21:12 ![Go to previous message Go to previous message](/theme/ovz3/images/up.png) |
dranch
Messages: 33 Registered: August 2007
|
Member |
|
|
Ok, so after rebooting the machine, I did the following:
VE#0:
/sbin/ifconfig eth1 0 #UPs the int and removes any IPv4 addr
/sbin/ifconfig eth2 0
/sbin/ifconfig eth3 0
/usr/sbin/vzctl set 100 --netdev_add eth1 --save
/usr/sbin/vzctl set 101 --netdev_add eth2 --save
/usr/sbin/vzctl set 102 --netdev_add eth3 --save
/usr/sbin/vzctl start 100
When I start VE#100, I now see VE#0's eth1 disappear and I can now "ifconfig eth1 up" in VE#100 as hoped! Woohoo... excellent news and it seems to work from initiate testing. This helps remove any MTU / VLAN tagging / etc. issues I might have seen if I had to deal with bridging virtual to physical interfaces.
Now, the issue of not supporting "eth0" in the VEs is unfortunate. Linux always enumerates from 0 so not having eth0 but say eth2 in a VE is both a change in behavior as well as just confusing to me and I'm sure to my developers as well. I'd love to see the ability to rename VE interfaces if possible.
--David
|
|
|