Open Stack

Scott Lowe <scott.lowe at scottlowe.org> wrote on 04/08/2015 09:56:40 PM:

> Please see my responses inline, prefixed by [SL].
> 
> Mike Spreitzer <mspreitz at us.ibm.com> wrote:
> 
> > Oops, my original email did not contain all of the constraints. Each
> > compute instance gets only one network interface, only one IP address.
> > Does Juno Neutron have a way to solve this problem?
> 
> 
> [SL] Use James' approach as described below, except with a single 
network
> interface on each instance. The OS inside the instance will see a single 
IP
> address (a private address from the tenant network), but a floating IP
> address can be associated with the instance. You'll probably need to
> implement split DNS (so that instances on the tenant network will 
resolve
> the address from the tenant network's subnet while external systems will
> resolve the floating IP address).
> 
> Even so, you could run into problems if the application running on those
> instances needs to report an IP address to other systems; it will report 
the
> private address, which may not be accessible by the other system.
> 
> 
> > Thanks,
> > Mike 
> > 
> > 
> > 
> >> From: James Denton <james.denton at rackspace.com> 
> >> To: Mike Spreitzer/Watson/IBM at IBMUS, openstack 
> <openstack at lists.openstack.org> 
> >> Date: 04/08/2015 07:04 PM 
> >> Subject: Re: [Openstack] [neutron] how to use multiple external 
networks? 
> >> 
> >> 
> >> 
> >> Hi Mike, 
> >> 
> >> With those requirements, I think dual-homing the instances may be the
> >> best approach.
> >> 
> >> In my mind, you would have 5 networks: 
> >> 
> >> A - External Network 1 
> >> B - External Network 2 
> >> C - Tenant Network 1 
> >> D - Tenant Network 2 
> >> E - Shared Tenant Network (No gateway) 
> >> 
> >> Because routers can only connect to one external network at a time, 
and a
> >> tenant network can only be connected to one router at a time, you 
would
> >> need two routers:
> >> 
> >> Router 1 
> >> Router 2 
> >> 
> >> You would connect them as follows: 
> >> 
> >> External Network 1 <-> Router 1 <-> Tenant Network 1 
> >> External Network 2 <-> Router 2 <-> Tenant Network 2 
> >> 
> >> The VMs would then connect as follows: 
> >> 
> >> Tenant Network 1 <-> VM1 <-> Shared Network 
> >> Tenant Network 2 <-> VM2 <-> Shared Network 
> >> 
> >> With no gateway set on the shared network, you won't have to worry 
about
> >> multiple default routes, nor do you need to worry about terminating 
that
> >> network off a router. It's simply isolated.
> >> 
> >> Hope that helps, 
> >> James
> >> 
> >> 
> >>> From: Mike Spreitzer <mspreitz at us.ibm.com>
> >>> Sent: Wednesday, April 8, 2015 12:28 PM
> >>> To: openstack
> >>> Subject: [Openstack] [neutron] how to use multiple external 
networks? 
> >>> 
> >>> Supposing there are two external provider networks, and a tenant 
wants
> >>> (a) some of his Compute Instances to have floating IP addresses on 
one
> >>> of those external networks, (b) some other of his Compute Instances 
to
> >>> have floating IP addresses on the other external network, and (c) 
all of
> >>> his Compute Instances to be able to talk to each other using only 
tenant
> >>> private networking, what arrangement of tenant networks and routers
> >>> would accomplish this? In Juno, if it matters.
> >>> 
> >>> Thanks,
> >>> Mike 
> 
> 

(sorry for top-posting the first time, I am switching to the usual 
convention now.)

Thanks Scott and James.  What I was trying to say about all the tenant's 
Compute Instances communicating with only tenant private networking is 
that the tenant does not want to have to use floating IP addresses for 
that intra-tenant communication.  After all, public IPv4 addresses are 
scarce; as an operator I am looking for a way that a tenant of mine can 
wire up all his compute instances without using external addresses for 
internal communication.  I guess I should have explicitly added to my 
original (a) and (b) a third category: some of the tenant's Compute 
Instances do not need to accept any inbound connections from outside the 
tenant's network (and thus do not really need a floating IP address).  As 
an operator, I am looking at the general case for how my tenants (which 
happen to be constrained in the one way I mentioned) can efficiently work 
in an environment with two external networks.

I was wondering if the following variant on James' topology would work: 
put the Shared Network between Router 1 and Router 2.  Is there a way to 
do that without even attempting to make Router 1 and Router 2 be gateways 
for the  Shared Network?  After all, the Shared Network only needs to pass 
traffic between Tenant Network 1 and Tenant Network 2, it does not need to 
handle any other traffic.

Thanks,
Mike

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20150408/9dba585e/attachment.html>