[ixpmanager] Notes on the install script [Re: Server rebuild]

[Barry O'Donovan]

Hi Rowan,

Just some notes on your install script (which, while I haven't had the 
time to look at it - yet - I also have absolutely no issues with 
whatsoever):

> (A) Follow the IXP-M wiki exclusively if you have lots of free time,
> and don't mind rediscovering platform quirks and having to recreate
> your workarounds each time

Or update the wiki with OS specific quirks for those that follow :D

> (B) Use it in parallel with reading the commands in the auto-installer
> script

Yes, this is a great source of secondary documentation.

> (C) Only actually run the autoinstaller (on a bare or fully backed up
> OS!) if you want to automate everything and you feel reckless, and don't
> mind hacking it a bit to fit with latest IXP-M versions

I can actually see this script being very useful for building 
development environments.

One thing some people may incorrectly assume from the installation 
script is that IXP Manager goes on /one/ server. This is not necessarily 
true and a bad idea in a production environment. At INEX we have:

  - IXP Manager installed on our web server for the front end;
  - IXP Manager installed on an admin server for executing all the 
command line / cron scripts
  - a monitoring server for MRTG / Nagios / Smokeping
  - a dedicated sflow server for p2p graphs
  - of course, the route collector, route servers, as112 service, etc 
are all dedicated servers / devices as well.

Take the p2p server as a good example. You need to pick a sample rate 
that doesn't create too large a quantity of traffic while also being low 
enough to create reasonable p2p graphs for the smaller trafficing 
members. As your exchange grows, this will require a lot of (a) 
processing power, and, also, as the number of members grow, (b) a lot of 
disk IO for all the rrd files. So, in a production environment, you 
really want to avoid an all-in-one installation.



More generally, running an IXP isn't easy. And, while we should rightly 
strive to make it as easy as possible and to ensure IXP Manager is as 
easy as possible, I'd be disappointed if people thought it could or 
should be boiled down to the simplicity of 'apt-get install ixp'.

Installing all the services around IXP Manager - like route servers, 
mrtg, p2p, etc - isn't meant to be a chore and one size definitely won't 
fit all. By doing these through the documentation:

  - you get to know and understand your own system;
  - you get a grasp of where and what data is stored (which will help 
plan upgrades and expansions);
  - when things start breaking, you'll have a better idea of where and why;
  - you also get to contribute back to the documentation so that these 
pages for items such as P2P / Mrtg / etc aren't just a step by step for 
IXP Manager but can be more useful explanation of "hows and whys" and 
even become a source of best current practices.


  - Barry


PS: For my own development environment, I tend to use a near clone of 
the production database and, via VPN, have access to the various MRTG / 
sflow servers and so have no need to recreate those locally.