openstack team mailing list archive

[Yan Zhai <zhaiyan920@xxxxxxxxx>]

Hi John,

   thanks for your careful answer. I just backed from vacation, too and
wish you have had a good time.

   I have been reading these documents again and again to check if I missed
something. My current understanding is: when I have two NIC, I should
configure xenbr0 and xenbr1 as default. Then under flat network assumption,
for each traffic type, I should create a vif in each control domU, with a
specific flat IP address range assigned to them (These addresses should
correspond to the bridge I choose to let flat condition hold). And when
each guest is starting, they will can have two vif, one for VM network, and
one for optional floating public address.
   Am I right about the configuration here?  If true, then is it also
combine the traffic of, for example, management and public access?  It may
sound stupid but I just want to confirm if it works as I expected.

   for multi-node nova-network I think I almost get it. I previously
thought each of which works individually but from nova-manage result I
think I was wrong. The only thing that still troubles me is whether I write
the correct config in nova.conf. I think 'xxx_interface' should be filled
with VIF name, and 'xxx_bridge' should be filled with BRIDGE name, is this
correct? Looks a little confusing anyway.

   I have manually patched the vif script with the 5.6 patch, but each time
I start the host-rule script, the VM is then not accessible from ssh.

   lastly for quantum, the major difficulties for me is the agent
installation of OVS plugins. It seems agents will rely on quantum modules,
which then try to load 'libudev'. However, libudev does not exist on CentOS
5.4 as far as I know, so that prevents me from marching on. Maybe this is
not the case for Debian/Ubuntu-XCP development, but it does trouble XCP
using.

thanks again for your help.

On Tue, Nov 20, 2012 at 10:59 AM, John Garbutt <John.Garbutt@xxxxxxxxxx>wrote:

> There are docs here:
>
> http://docs.openstack.org/folsom/openstack-compute/admin/content/introduction-to-xen.html
> And networking info here:
>
> http://docs.openstack.org/folsom/openstack-compute/admin/content/xenapi-flat-dhcp-networking.html
>
> Let me know what bits are confusing, and I will make an effort to improve
> the docs around that area, and help you through things.

Firstly, quantum support in XCP (with OVS) is still a work in progress.
> Help to review the changes in Gerrit are very welcome!

Now, lets look at your questions...
>
>  > 1 Controller node: no Xen, only Ubuntu 12.04, everything for openstack
> service except for nova-compute
> > 2 Compute node: XCP 1.6 beta, with nova-compute in special domU (Ubuntu
> 12.04), xenapi plugin installed in dom0
> > each node has two NIC, one with public IP (Only limited floating IP),
> and another in private network (Any IP is OK)
> > flat network or flat dhcp network
> > I want to use eth0 for public traffic and service request, and eth1 for
> inter-vm traffic.
>
> OK, but where is your management traffic, like Rabbit and MySQL going? I
> would have expected one nic for management network, and one nic for public
> traffic, with maybe with a separate VLAN for instance traffic?
>
>

> 1.) does each domU need nova-network running? My understanding is it's OK
> to run nova-network individually, but then how to mange the floating IP
> globally?
>
> Correct. If multi_host=true, you are running nova-network on every node,
> if false you just run one. It depends if you don't mind the extra overhead
> or not. Others can describe the trade-offs more clearly.
>
> http://docs.openstack.org/folsom/openstack-compute/admin/content/xenapi-flat-dhcp-networking.html
>
> 2.) in document for flatdhcp network, I saw four interfaces for each
> management domU. Is it OK to have only two interfaces? Say, eth0-xenbr0 for
> public IP and services, and eth1-xenbr1 for VM network?
>
> As above, you need to think about where your management traffic like MySQL
> and Rabbit traffic is going. Three is certainly fine.
>
> 3.) Is the network isolation rules a must for test install? I found the
> patch to vif is still for xenserver 5.6_p2, and can not be applied to vif
> of xcp 1.6 or xenserver 6.1, which might be a trouble.
>
> It is not required for a test setup. Do bug me if you are having problems
> with that stuff on XCP, and I can ask the XCP guys to help me get that
> working for you.
>
> Sorry for the slow response. I was on holiday.
>
> Thanks,
> John
>
> > From: openstack-bounces+john.garbutt=citrix.com@xxxxxxxxxxxxxxxxxxx
> > [mailto:openstack-bounces+john.garbutt=citrix.com@xxxxxxxxxxxxxxxxxxx]
> > On Behalf Of Yan Zhai
> > Sent: 20 November 2012 7:52 AM
> > To: Afef MDHAFFAR
> > Cc: openstack@xxxxxxxxxxxxxxxxxxx
> > Subject: Re: [Openstack] Networking issues for openstack on XCP
> >
> > have even more trouble setting up quantum itself. The plugin support in
> > XCP looks not as good as KVM: the first problem is python 2.6 and it's
> solved
> > by adding new repository. But when I installed the openvswitch agent, it
> > can not start, requiring some quantum python module, which then seems
> > have dependency on libudev. And this means I have to update almost the
> > whole dom0, including libc, udev, and blablabla to get that library.
> That's
> > too risky. Maybe I shall stay with bridge plugin and see if things can
> be a
> > little better...
> >
> > On Sat, Nov 17, 2012 at 1:18 PM, Afef MDHAFFAR
> > <afef.mdhaffar@xxxxxxxxx> wrote:
> > Hi all,
> >
> > I am also trying to install openstack, with xcp. I used devstack to do
> that,
> > since it is more simple.
> > However, I am still facing a network problem.
> > Actually, I got a private network per physical node. I am able to access
> the
> > tenant VMs from the corresponding openstack DomU. But, these tenant
> > VMs are not inaccessible from any other machines.
> > Is that normal (ie. a private network per physical node? The created
> tenant
> > VMs can access external machines, but they are invisible to other
> machines?
> > Is there any way to let my VMs accessible, at least from other tenant VMs
> > (created on other physical nodes)?
> > How can Quantum solve this problem?
> >
> > Thank you
> >
> > Regards,
> > Afef
> >
> > 2012/11/17 Yan Zhai <zhaiyan920@xxxxxxxxx> Hi Robert,
> >
> >  thanks for reply. Currently I am just looking for a way to bring it up
> for
> > internal trial, so if Quantum is better I will move to that. The only
> reason
> > that I am still asking for questions about nova-network is because of the
> > document order: I am setting things following the install document, but
> > when it comes to the network part I encountered above confusions. I will
> > check the quantum document to see if anything can be simplified. Thanks
> > again!
> >
> > best
> > Yan
> >
> > On Fri, Nov 16, 2012 at 11:54 PM, Robert Garron
> > <Robert.Garron@xxxxxxxxxxxxxx> wrote:
> > Yan,
> >
> > In my opinion, if you are going to spend all the time learning a new
> product
> > -- i.e. nova network vs quantum.  And if you are only testing a concept,
> I
> > would spend it upon Folsom and move from Essex and/or nova network and
> > nova storage.  Quantum eases many of the issues Nova network has  or had,
> > but quantum is the system for support and deployment going forward....
> > just my 2 cents...
> >
> > Robert
> >
> >
> >
> > On 11/17/12 12:33 AM, Yan Zhai wrote:
> > Hi,
> >
> >    I am trying a proof of concept install of openstack on xen cloud
> platform.
> > However, I am a little confused about the network part in Folsom.
> >    My architecture:
> >      1 Controller node: no Xen, only Ubuntu 12.04, everything for
> openstack
> > service except for nova-compute
> >      2 Compute node: XCP 1.6 beta, with nova-compute in special domU
> > (Ubuntu 12.04), xenapi plugin installed in dom0 already
> >      each node has two NIC, one with public IP (Only limited floating
> IP), and
> > another in private network (Any IP is OK)
> >
> >    My goal for network:
> >      flat network or flat dhcp network. I want to use eth0 for public
> traffic and
> > service request, and eth1 for inter-vm traffic.
> >
> >    My questions:
> >      1.) does each domU need nova-network running? My understanding is
> it's
> > OK to run nova-network individually, but then how to mange the floating
> IP
> > globally?
> >      2.) in document for flatdhcp network, I saw four interfaces for each
> > management domU. Is it OK to have only two interfaces? Say,
> >           eth0-xenbr0 for public IP and services, and eth1-xenbr1 for VM
> > network?
> >      3.) Is the network isolation rules a must for test install? I found
> the patch
> > to vif is still for xenserver 5.6_p2, and can not be applied to vif of
> xcp 1.6 or
> > xenserver 6.1, which might be a trouble.
> >
> >    If anyone could help me kindly, it will be quite appreciated, I have
> been
> > stuck with install document for several days.
> >
> > thanks a lot!
> > - Yan
> >
> >
> > _______________________________________________
> > Mailing list: https://launchpad.net/~openstack
> > Post to     : openstack@xxxxxxxxxxxxxxxxxxx
> > Unsubscribe : https://launchpad.net/~openstack
> > More help   : https://help.launchpad.net/ListHelp
> >
> >
> > _______________________________________________
> > Mailing list: https://launchpad.net/~openstack Post to     :
> > openstack@xxxxxxxxxxxxxxxxxxx Unsubscribe :
> > https://launchpad.net/~openstack More help   :
> > https://help.launchpad.net/ListHelp
> >
> >
> >
> >
> > --
> > - Yan
>



-- 
- Yan