Resources Mango needs
MattFox last edited by
Just trying to cover my angles here. But is there an actual rule you go by for how many cores and how much ram is advisable for mango to use depending on the number of points, datasources, and threads?
I'm half wondering if issues I have had are resource related.....
I also had problems with our gateway to the server. Look at the story
It is very helpful if you activate all internal data points in the data source "Mango Internal" and make a dashboard to watch them. Only then you have an overview of what leads to which problem.
The cron-job function can also help to distribute the load evenly. Not all updates (data source and ptp) at the same minute of the hour ...
So, for our 32000 data points, 268 data sources and 10 ptps, syncing history every 5 to 15 minutes, setting max 5 GB RAM I think this hardware is the minimum:
If you add some webclients you can overload the system very quick, but it's only a gateway.
The ideal computer is still to be developed. You enter your problems and they never come out.
MattFox last edited by
@ralf Many thanks, I'll be sure to give it a go.
how many points and datasources at what poll rates and logging are you running on what hardware?
We had one mangoES die after about ~2 years of service with less than 100 points logging once per second, but up until then it seemed to do OK.
MattFox last edited by MattFox
6 Core xeon 15GB server
Given Mango 10G to run with, htop seems to show it asking for 14 in a virtual sense, not sure if that means I need to crank it up or not..
377 Datasources, A little over 8200 points, About half of those being updated by API from a third party system. All points are updating ranging from 20 seconds to 15 minutes. I'd argue about 500 points are meta points.
193 users, I've always got about 2/3 thirds having active sessions on the system.
CPU hits approx 20% every once in a while. I've got 700 max threads running in settings to ensure mango can just smack everything out as it needs. But after having issues last week with tags being saved to show status information I was beginning to wonder if there's more I'm missing....
cwangv last edited by cwangv
I want to give you a comparison from the point of view of Building Automation Systems (BAS) that will have to deal with such massive amount of points.
Generally, the BAS will have network controller which is similar to a MangoES that polls various devices then either do something with them or perform data-sharing between network controllers.
To accommodate 32000 points and maintain acceptable performance, 2000 physical/software points from all these devices is generally the limit i will put on each network controller.
So, i will need at least 16 network controllers. So that will be 16 CPUs (or cores) (one per network controllers, they could be ARM or similar, keep in mind, these network controllers CPU are much slower than a mid-range desktop CUP). Each network controller may have 2GB RAM, so there will be 32 GB RAM required.
In addition, in system this big, there will be a separate Server-Class PC to serve up all the webpages to the end-users. This server-class PC will at least have 8-core and 16GB RAM. All up, there are at least 24 cores CPUs and 48GB RAM required to run this system.
Looking at your server spec, it is under-powered.
I hope the above makes sense.
I agree with you but our Mango-PC on side is only a data collector and gateway via Persistent TCP Publisher to the cloudserver. The Mango-PC has no webpages to serve for end-users. This is the job for the cloud-server.
Our 268 data sources are 50% BACnet and 40% SNMP and 10% Modbus TCP/IP.
All BACnet-Devices are BACnet Building Controllers (TCP/IP), propper setup with COV etc, minimum network traffic.
The installation is running since 9 years with an other software as data collector and gateway, and now with Mango.
If I want to serve the webpages on side, then I have to install a much larger system as server and data-collector, I know.
MattFox has 127 users to serve, a complete other situation. A web server process with this size needs a lot of RAM and core-power.
The new experience with Mango was there are a lot of adjustment screws and you can watch and tune the system very fine, and you have to do it :)