Starting somewhere in 2006 I've rented a 1U rackspace from TransIP. They were the cheapest colocation company in the Netherlands I could find then, and they still are. Costs have risen in the mean time though, mostly because of power usage. When I started, the server was an old Dell PowerEdge with 2 Pentium III's at 800Mhz, 1GB ECC RAM (4 x 256MB sdram) and 2 15GB 10k SCSI disks. I installed the then latest Fedora Core 4 on it. Since then I've hosted various websites and email domains on it, when in 2009 2 memory modules went dead. I started thinking about replacing it. One of the things that bugged me was that it was virtualy impossible to upgrade the OS without extensive downtime and/or long hours at the datacenter. In the mean time, by the end of 2010, I had gained more experience at virtualisation, especialy bare metal hypervisors. The PowerEdge's hardware however was nowhere near capable of running ESXi 4.1, which was the hypervisor I chose to work with. So I set out to find some hardware somewhat compatible of running ESXi.

In Februari 2011 I found and bought a second hand server system from Intel, a SR1425BK1-E. I expanded the memory to 4GB and put in 3 80GB Sata disks. Then I installed ESXi 4.1 on an USB disk and after setting up the required virtuals (router/firewall, mail/webserver and a backup Windows vm to run vSphere client) I installed it in the datacenter and put the old Dell offline.

When the need would arise to replace the OS of the virtuals I would simply setup a new one and migrate what ever needed to be migrated. But the new Intel server system had some limits, for some reason it only gives 3.2GB RAM to the hypervisor and the hypervisor is limited to hosting 32bits virtuals because of the hardware. All in all it was still a major improvement. After a month of operations I was presented however with a bill for power over usage. 86% to be exact...

I chose to except the additional costs, they were somewhat managable. In Januari 2012 my contract's initial periode expired though and the basic price was raised. This was a bit to much to keep paying and I set out to replace the Intel system with something with less power hunger.

Intel has some very interesting CPU's with a very low TDP, especialy in comparison to the 135Watt TDP of the 3.2Ghz CPU in the server system. I calculated I had about 85-88Watt to fill before I would have to pay for additional power usage. A 65 Watt cpu seemed to be cutting it to close, the mainboard and disk(s) would also need power and I didn't want to end up on the wrong side of the 85Watt mark. VM's mostly use memory, there's no high need for CPU power. So I decided on an Intel Pentium G630T with a TDP of 35Watt. A Gigabyte mainboard, 2 x 4 GB DDR3 and a 500GB baracuda ES would complement it. Next was the housing, this presented an additional challenge. 4U housings are plentifull, even 2U's are around in numbers. 1U's are different though, 4,5cm hight isn't much to work with so they present some challenges. After a few months of saving money and searching the internet I settled on a Supermicro Superchassis 502L-200B. I checked if it could house the mATX mainboard I chose and I found a 28mm high CPU fan.

When I tried to install the mainboard though I saw that the backplane was not compatible and neither was the PSU. Only SuperMicro Atom server boards are compatible! So nice to see no warning at all at the website or in any of the reviews. I made a quick search for alternatives but they all exceeded my budget. So I set out to adapt the housing. I found a 1U PSU and I broke out the backplane. Et voila, it works!

Next up was the new hypervisor. ESXi 5 was out a year by now, so I chose to go for it. I knew the onboard NIC was not on the HCL so the first thing was to install the necessary drivers in ESXi. That was pretty straight forward though with ESXi-Customizer-v2.7. Then I set up the first virtuals. A new pfSense firewall/router, a CentOS 6.3 Webserver/MySQL server, a seperate CentOS mailserver and a backup windows vm. But of course 2 days before I had planned to install the new host a pretty major issue came up. the free edition of ESXi 5u1 has a misconfiguration and doesn't auto power any virtual after a (re)boot. Well, just connect vSphere client and boot them manualy. I only have 1 IP and it's assigned to the firewall's WAN of course. And it doesn't power on.. Catch 22. But thankfully there's a workaround for the moment. Just add a script to /etc/rc.local that will use CLI commands to boot the virtuals anyway.

The new host is in production now for about 4 days, so I won't say to much about it except for it's powerusage. It dropped from 744mA to 170mA, which corresponds to about 37,4Watt. Nice!


Share this post

Design by JoomlaSaver