Results 1 to 1 of 1

Thread: Distributed monitoring with several layers

  1. #1
    Junior Member
    Join Date
    Oct 2015

    Distributed monitoring with several layers


    I'm trying to come up with the best way to deploy Shinken under what I imagine are some pretty unusual constraints, and I'd just like some feedback on my thoughts as they stand now before I go spending hundreds of hours on something that turns out to be fundamentally broken.
    Our architecture consists of a couple of layers, both in terms of network topology and physical locations, I'll try to make up sensical names for them and explain briefly:

    DC layer
    The "master" layer at the top of the hierarchy. This In turn has a few "sub layers" but I don't think they'll be terribly important here since this is essentially just your average DC as far as networking goes.

    Zone layer
    This sits in the same physical location as the DC layer and is essentially the glue between the DC and Site layers, and manages the Sites. It's got connectivity to the sites below, and it is in turn managed from the DC layer.
    There is one "replica" of this layer for each zone, a zone in turn consisting of any number of Sites(could be anything from just 1-2 up to a hundred or more Sites). For simplicity we could call this a country, there's a bit more to it than that but I think from Shinken's perspective it's close enough.
    There will likely be something like 50+ of these.

    Site layer
    These are sites spread all across the world, with varying quality of connectivity to the Zone layer. The site layer essentially has two sub layers, might as well call them Server and Client to keep the terminology simple. All of it is still managed by the Zone layer though.

    So I hope that makes sense, it's a fairly basic hierarchy. DC -> Zone/Country -> Sites

    Now, that out of the way, the arbiter(s) will sit in the DC layer, I don't see any other sensible way to do that since traffic will have to flow downwards through the layers. Right now my idea is something like this:
    DC - Arbiters, brokers, reactioners

    Zone - One realm per zone, will have both schedulers and pollers.

    Site - Preferably no logic here. If need be stuff like an NRPE agent or some such.

    This makes the most sense to me since the DC layer will be able to talk to the Zone layer, and the Zone layer in turn will be able to talk to everything in the Site layer. And in terms of "business impact" the Site is pretty much a unit in itself, rather than a bunch of components.
    And of course there's the issue of scale, as you might imagine this will grow to quite a few units, depending on exactly what we decide to monitor and with what granularity, it might grow to tens of thousands of units and god knows how many checks we need monitored, though I'm always leaning towards keeping it as compact and simple as possible.

    Sorry for being long winded, but I think this is the shortest I can manage without leaving out essential stuff. Any takers?
    Last edited by Sunner; 10-19-2015 at 04:10 PM.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts