I spent last week (the last of my three week trip to Europe) in the German city of Köln (Cologne). I’ll probably keep referring to it as “Köln” since I just figured out how to easily print special characters on my Ubuntu desktop. (grin)
I was working with a service provider called NetCologne, specifically on a project where they are rolling out a Fiber to the Building (FTTB) project. Ronny Trommer, one of our guys in the European office, did most of the heavy lifting, but I was there to help out (at least in the beer drinking capacity).
We needed to monitor and collect data from a growing number of remote devices, and when I last checked we were polling 26110 distinct devices and 51690 interfaces. One thing that came in handy with this deployment was the Instrumentation Log Reader.
If you haven’t seen this feature, you should check it out. Under Admin there is a menu item called the “Instrumentation Log Reader”. This will generate a very useful report from the instrumentation logs concerning how many devices are being scanned for data collection, how long it takes each scan to complete, the time period between scans and the number of failures.
At NetCologne with 300 threads (the default is 50) the system was not able to poll fast enough to provide five minute samples, so we needed to increase the number of threads. While it was not easy to do a weighted average, it seemed that a conservative estimate would be about 10 seconds per device to perform collection. There were some of them in the 20-30 second range, and quite a few below 5 seconds, but for calculation purposes 10 seemed like a good number.
This meant that each thread could collect on six devices a minute, or 30 devices per five minute period. Since we needed to collect on 26110 devices, that meant we needed a little over 870 threads. We made it 1000 just to be sure, and then data collection started performing well.
A lot of the time was spent integrating their internal systems with OpenNMS through the provisioner. Much like Towerstream, any service provider with this many devices needs to be able to automate adding and removing devices to their management system. I am certain that the strength of our provisioning system is the one major reasons large enterprises and carriers should consider switching to OpenNMS.
I found out on Google Plus that a number of people have very strong opinions about the quality of that particular beer, but I liked it. It was very light and clean, which was helpful in that we drank quite a bit of it. It came to our table in a “ring” of glasses:
and we went through several rings (I lost count).
No matter what you may think of kölsch, I don’t think anyone can deny the beauty of the cathedral:
Even at night covered in part by scaffolding, it is a gothic, fractal masterpiece.
I had fun, and I look forward to the opportunity to go back as this project progresses.