Not bad for a first effort…
So I am given control of a dual quad core Xeon server (i.e., 8 cores in all) with 2TB + 500GB Hard Disk arrays, plus 32GB RAM. Not the best box we’ve got in the work server room, but pretty nice all the same. Out comes my newly-minted VMware ESXi Server installation CD, and within minutes, the installation is complete. It takes me a further four days (plus an all-nighter), but I finally ended up this afternoon with 7 perfectly-formed Windows 2008 x64 virtual machines running all our test and development Oracle environments. Those same seven databases had previously lived on at least three servers, but things got moved around so much it was hard to tell. The server copes admirably with the load (well, the developers and testers can’t be said to hammer the things hard, I suppose!), and a 7-to-1 consolidation (or even a 3-to-1 ratio, depending on the way you look at it) seems a good outcome all round.
A couple of niggles (inevitably). First, one of our arrays actually consists of 6 500GB disks. We use RAID-5 (and I’m prepared to argue strongly that doing so makes perfect sense in a lot of environments, so let the BAARF people commence battle if they must!), so that means there are approximately 2350GB of usable space on the array… yet ESXi reported it as a datastore of precisely 512 Bytes. What’s a few thousand missing Gigabytes between friends?! Turns out that ESXi 4.0 has a 2TB limit on any one datastore, which meant I had to re-configure the array as a 5-disk one with a hot spare: 1835GB and therefore under the magic number, and accordingly reported correctly (and entirely usable as a result).
Second, in order to make sure my home network resembles my work one as closely as possible for R&D purposes, I attempted to install ESXi on a dual-core box I have spare (only 4GB RAM, but a couple of terabytes of usable disk space, so nothing too bad). Sadly, the installation died an abrupt death with a Failed to load lvmdriver error and continued to do so, no matter which of my many spare Netgear cards I installed, or whether I stuck with the on-board Realtek network interface. The point being that ESXi 4.0 uses the network card to generate a unique ID for each host server -so if it doesn’t support the NIC you’re using, the installation can’t proceed. Nothing that an AU$40 purchase of a proper Intel PRO/1000GT network card wouldn’t fix, however -so things are running nicely here, too.
The whole thing is managed by the VSphere Client (free, and downloadable from the server itself, once it’s been installed properly). It’s a lovely GUI tool that makes creating, snapshotting, performance monitoring and generally managing multiple guest VMs a piece of cake.
I am very enthusiastic about the final outcome at work (especially considering the juggling I had to do to get thereĀ -moving databases all over the place multiple times to free up the server, so it could be rebuilt, then moving them all back again… and all without disrupting important testing and development efforts going on at the moment, for example). I was, of course, enthusiastic about XenServer too, before I bumped into its own bunch of quirks and ‘features’ that proved show-stoppers. Time will tell if the same is true for ESXi… but it’s not looking too bad at the moment!
October 30th, 2009 at 11:07 pm
Hi.
I’m sure once it stabilizes all the people at work will see the benefits and go VM mad.
I remember doing server consolidations using VMware GSX Server running on RedHat 7.1 (yes, pre Enterprise Linux days). Even then it worked like a dream and after the initial skepticism everyone, including me, became converts.
Good luck.
Cheers
Tim…
November 4th, 2009 at 7:07 pm
Funnily enough, the two system admins are now installing ESXi and virtualising like they invented the concept. But I’ll remember who got it in the door first, even if they choose to forget!