Robert V., It may be better to examine this from the eco system point of view, since building a comprehensive suite from scratch is not trivial.
We have 2 ecosystems: Telco and Enterprise. Telco is moving towards an ETSi like implementation. So, lets place our bets. I am going to bet that AT&T will not allow ONAP to fail. So we need to get together with ONAP on monitoring feed standards and implement to their spec. Enterprise We are heading towards a container based world while at the same time getting a lot of our current traction from openstack. Between Prometheus and Monasca I would bet on Prometheus, I think decreasing software spend on OpenStack will doom Monasca. So lets create a Prometheus data source and add display capabilities to it. For a while maintaining a stand-alone solution in the community makes sense, we have an existing one, reinventing it would not serve us well. Longer term we should probably consider on how to interface well to both pull and push based monitoring systems and create referenceable designs on how to do this. Not clear to me that intergrating with low level tools like collectd makes sense. I do not think that there is that much demand for home grown monitoring systems Appformix is a single pane of glass monitoring system going from infra to app level. That is the direction Juniper corp is taking, evolving from SDN specific monitoring to a comprehensive monitoring solution, look at it as an expensive validation that Robert R. is quite correct in his analysis ☺ Greg From: "Van Leeuwen, Robert" <rovanleeu...@ebay.com> Date: Sunday, September 24, 2017 at 11:35 PM To: Gregory Elkinbard <gelkinb...@juniper.net>, Robert Raszuk <rob...@raszuk.net> Cc: "dev@lists.opencontrail.org" <dev@lists.opencontrail.org>, "krzysztof.klimo...@codilime.com" <krzysztof.klimo...@codilime.com> Subject: Re: [opencontrail-dev] [TSC] Analytics and web ui on OC 5.0 and beyond Sorry about the previous empty message. Keyboard-shortcut booboo. > To your comment on analytics it is a mistake to think of Contrail or Nuage or > Calico as detached solutions. > They are part of entire end to end network and just like it is today with WAN > routers, with DC underlay, > with Internet ASBRs in any company the analytics should be exported and > processed to central repository and management system for the *entire > network*. With this train of thought you can argue that the network is part of something bigger than just the network itself. IMHO organizations will already have something in place for analytics/alerting which are not a “side-project” of a, in this case, network solution. e.g. in our case we would also want to see the application landscape in that same view of the network. And that’s currently one of the many reasons we do not look at analytics: it is another place to look at stuff. We would want something that, at the least, can be displayed by a grafana dashboard together with all the other graphs. e.g. since everything else is in a graphite cluster we are parsing output of dropstats and the introspect into graphite just to get everything in one place. Which is not exactly optimal but it is very convenient to have graphs of a year back without: - looking at another system - dealing with access rights in contrail to get everyone view access (e.g. grafana makes read only access for everyone very easy) - cassandra database explosion and other analytics (scaling) issues So what I would like to see is some easy integration with the best tools for the jobs already out there. Just to name a few things many companies are using anyway: collectd metrics collectors / graphite protocol / SNMP those kinds of things. As long as metrics/monitoring solution X is easily integrated I would be happy. Cheers, Robert van Leeuwen
_______________________________________________ Dev mailing list Dev@lists.opencontrail.org http://lists.opencontrail.org/mailman/listinfo/dev_lists.opencontrail.org
