I am working on enhancing a BI platform, plugged into a bunch of stats gathered from a virtual datacentre (based on Jasper, but this shouldn't really make a difference).
So far, I have 4 hypervisors and a few TB of storage, so I can bring up probably up to 100 (more or less) beefy VMs.
what I want to do is basically simulate a heterogenous DC with different Linux and Windows servers that are puffing away at their daily routines, with monday morning effects and incidental peaks, and idle time too of course. All that, to gather enough pseudo-live data in the BI server, so I can start working on what data is missing, what reports make sense to develop and so on.
I would like to gather some opinions on how to simulate a real working DC, from what services I would require, to how to properly stress them, and what to expect from these stress tests (so I can verify it in the stats later on)
So far I have an email server, directory (AD and IPA), DHCP, monitoring (nagios). I will want a couple of DB servers (postgres and MSSQL probably), maybe a dozen VDI desktops of various kinds...
Any ideas you throw my way would be great, both in terms of the servers that it would make sense to have in this demo lab, and in terms of how to stress them in a measurable way.
You can simulate SQL load by taking a production DB and getting a backup, then recording all of the queries over the day. In you fake data center, restore the database and replay those queries are the proper times to simulate real SQL server load.
You probably need some web servers (apache/IIS/nginx), ftp clients/servers (ftp, lftp, vsftpd, proftpd), ssh clients/servers (ssh, scp, openssh), misconfigured (open) proxy servers (squid, apache, tinyproxy), db servers (mysql, postgres, mssql), email servers (sendmail, postfix, etc - obviously sending huge amount of spam), torrent clients (rtorrent, utorrent), dns servers (bind), security tools (nmap) and some abusers to simulate a real DC ;-)