On Tue, Jul 11, 2017 at 8:55 PM, joehuang <joehu...@huawei.com> wrote: > Hello, operators, > > If you have moved mission critical application to OpenStack cloud, it's > running in one region or multi-region? Or if you plan to move your mission > critical applications to OpenStack based cloud, what's your proposal to > deploy the mission critical applications? > > During OPNFV Beijing Summit, we have done a demo to deploy both vIMS(telecom > application providing VoIP etc) and JITSI(video conference) in multiple > OpenStack clouds, both applications achieve cloud level high availability in > multi-region OpenStack cloud: > > 1) For vIMS, the application is active-active in two OpenStack cloud, the > backend db Cassandra will sync the data through internal cross OpenStack L3 > network (which was automated by OpenStack Tricircle project), the vIMS > service can be accessed through FIP in each OpenStack cloud, and can see > same data set. If vIMS is stopped in one OpenStack cloud, services can be > continued in another OpenStack cloud. > > 2) For JITSI, the video bridges are running in two OpenStack clouds, and the > heart-beat for video-bridges / control information to video bridges are > through internal L2 network (which was automated by OpenStack Tricircle). > Similar like vIMS, video streams for video conference will be handled by FIP > in each OpenStack cloud respectively. If video-bridge is stopped in one > OpenStack cloud, services can be continued in another OpenStack cloud. > > I also found that similar deployment use case happened in AWS. So I wrote a > blog to compare the deployments with what happened in AWS: "Doing better, > "Moving Mission Critical Apps from One Region to Multi-Region" > https://www.linkedin.com/pulse/moving-mission-critical-apps-from-one-region-joe-huang > > > Is this kind of deployment will be your favorite regarding to mission > critical applications? > > Today I read the thread "Tricircle" : > http://lists.openstack.org/pipermail/openstack-operators/2017-July/013914.html > , therefore I would like to open a new title thread here to bring broader > concerns and discussion for this topic.
I think this is a big topic. I was hoping some other people would comment. :) What Sony did with AWS is interesting, and I'd have some questions about their choices (like I wonder why they did straight up EIP over public internet as opposed to say a direct connect, maybe straight EIP is faster or more reliable...I've never had access to direct connect so I don't know). Overall, as someone working in the NFV area, I'm on calls all the time discussing making applications highly available over multi-site openstack deployments. I feel like it's really important. Presumably this means having an API that can connect tenant networks together is important, and possibly either anycast or somehow migrating IPs across DCs. I've been using the neutron RBAC stuff a bit lately and it is great to be able to share networks in a site between tenants (eg. a service tenant and a normal tenant), and would be fascinating to do that over multi-sites, but of course I'm leaving out a lot. Typically people hope that that is somehow taken care of by the "management and orchestration layer" (MANO) + the software defined network (SDN) system , but that is moving a lot of the work outside of OpenStack into APIs that aren't as standardized. Thanks, Curtis. > > Best Regards > Chaoyi Huang (joehuang) > > _______________________________________________ > OpenStack-operators mailing list > OpenStack-operators@lists.openstack.org > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators > -- Blog: serverascode.com _______________________________________________ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators