Linux serverNETWORK ADMINISTRATIONSredhat

How we designed an OpenStack Public Cloud at Scale with Fabric Networking

Our engineers at Nipa Cloud, the only OpenStack-based public cloud in Thailand, have learned what is needed to improve to compete with Hyperscalers. We are currently deploying a new cluster with multi-AZs and Victoria OpenStack integrated with Tungsten Fabric (TF) and new UX/UI portal for both self-service provisioning and pay-as-you-go billing.

We will discuss how we design OpenStack and TF architecture to support multi-AZs, and how we use the concept of Nova cell for scaling compute nodes. We will dive into how to design Ceph with multi-AZs. We will show how Glance uses S3 object storage as a backend with symmetric sync policy and Cinder with image volume cache.

In order to scale compute and storage nodes, we design underlay network as CLOS topology (Spine/Leaf) with BGP EVPN/VXLAN and Anycast Gateway. For overlay networks, we use TF to enhance network features such as Service Chaining which supports ECMP by forwarding data efficiently via MPLSoUDP.

Speakers:
Dr. Abhisak Chulya
Charnsilp Chinprasert
Jeeraput Supaapinun

Thank you to the Sponsors of the OpenInfra Summit Berlin 2022!

Connect with us:
OpenInfra Twitter: https://twitter.com/openinfradev
OpenInfra LinkedIn: https://www.linkedin.com/company/open-infrastructure-foundation
OpenInfra Facebook: https://www.facebook.com/openinfradev
OpenInfra Foundation Website: https://openinfra.dev/

#OpenInfra #OpenInfraSummit #OpenStack #OpenSource

source

by OpenInfra Foundation

openstack