Rob - sorry for the delay on this. Here’s what we are working on for an
architecture with our physical network environment (the so-called “under
network”) ...
Scale is based on a rack unit of servers. To easily manage scaling L2
boundaries (broadcast domains), we confine all L2 boundaries to the
physical rack, and the ToR is an L3 separation between racks. This allows
us to scale N-number of racks without concern to L2 boundaries. It also
allows us through standard protocol management (OSPF and BGP) utilize
*any* vendors network product; without using proprietary solutions like
meshes/fabrics (eg Cisco’s “Insieme”). No Spanning Tree L2 problems, we
get full active/active link connectivity from ToR (leaf) to aggregate
(spine), and the leaf nodes can scale out based on the chassis
size/capabilities (port count) of the spine switches. We can easily scale
our oversubscription rate out of the ToR to Spine via the number of
northbound links from ToR to Spine, retaining full active/active
capabilities.
A simplified example of this at the rack level and how that impacts the L3
topology of OpenStack deployment is as follows.
(“A”, “B”, “C”, and “D” refer to the L3 network in each rack for the
various OpenStack networks)
A = PXE/Admin boot network
B = OpenStack management/API network
C = OpenStack private (SDN)/storage network
D = OpenStack public (floating etc.) network
rack 1 rack 2 rack N …
A = 1.1.1.0/24 A = 1.1.2.0/24 A = 1.1.N.0/24
B = 1.2.1.0/24 B = 1.2.2.0/24 B = 1.2.N.0/24
C = 1.3.1.0/24 C = 1.3.2.0/24 C = 1.3.N.0/24
D = 1.4.1.0/24 D = 1.4.2.0/24 D = 1.4.N.0/24
Transit between Rack 1 to Rack 2, or Rack “N” for that matter, is via L3
convergence at the Spine
Does that help illustrate what I'm thinking about ?
~~shane
On 1/7/14, 1:59 PM, "[email protected]" <[email protected]>
wrote:
>Shane,
>
>Interesting. I’m trying to imagine the topography because CB2 does have
>the concept of a network with multiple IP ranges assigned. Can you give
>a concrete example w/ IP subnet and function that you’d split?
>
>I’m not sure how you’d pick which range was tied to a L2 boundary for
>assignment.
>
>Would IPv6 help solve this?
>
>Rob
>From: crowbar-bounces On Behalf Of Shane Gibson
>Sent: Tuesday, January 07, 2014 1:26 PM
>To: crowbar
>Subject: [Crowbar] L3 topology deployment with Crowbar?
>
>
>
>
>
>
>
>All,
>
>
>
>I have a question regarding Crowbar deployment. Currently - I have
>deployed Crowbar 1.x in a flat L2 physical topology. Meaning - all of
>the OpenStack cluster members are in a single L2 domain.
>
>
>
>We are working on a physical datacenter topology where every single rack
>is an L2 boundary. Inter-rack traffic (east/west) is an L3 boundary. We
>will be utilizing OSPF and BGP to provide inter-rack redundancy and HA
>capabilities (I.e. no spanning tree between racks for failure).
>
>
>
>My question - can Crowbar network barclamp support multiple L3 network
>segments for a single cluster deployment? I don’t recall any
>documentation or designs that point to this being a capability (in 1.x).
>
>
>
>If that’s the case - does OpenCrowbar 2.x incorporate the capability in
>the design to support deploying a cluster across different L3 networks in
>a single cluster?
>
>
>
>Thank you,
>
>~~shane
>
>
_______________________________________________
Crowbar mailing list
[email protected]
https://lists.us.dell.com/mailman/listinfo/crowbar
For more information: http://crowbar.github.com/
