I finished installing, cabling, and labeling the new servers from iXsystems that we bought a few months ago.

They are now running FreeBSD 8.2.

In a previous post about attending the Agile 2010 conference, I included some quotes from Arin Sime who presented a talk on using range estimates rather than single-point estimates:

  "Single point estimates are almost always overly optimistic."

  "There is a limit to how well a project can go, but no limit to how many problems can occur."

There are many examples out there to show that single point estimation is hard and in fact most people are pretty terrible at it.  See the "Further reading..." section at the end of this post if you are in doubt about estimation woes.

I was able to convince our team to use range estimation rather than single point estimates in our most recent team project. The project in question was building a small datamart that will become part of our larger data warehouse efforts. We received a data feed from an external source and processed it through ETL (Extract, Transform, Load), with most of our work in the "transform" portion of the process. The deliverable was a series of reports on our web site and we now have the data loaded for any future ad-hoc queries.

 

In our project plan we recorded three values for each task:

  1. Minimum - this would be most people's initial single point estimate

  2. Maximum - maximum reasonable value if things went really wrong with this task

  3. Two-Thirds method - the calculated point two-thirds of the distance between the minimum and maximum.

 

To calculate the two-thirds point:

Today I replaced the old UPSes that had been powering our server rack... well, I should say that I replaced the dead one as well as its brother who was still hanging on for dear life.

This is the vintage of old UPS I removed:



We replaced the old units with APC Smart-UPS 2200VA SUA2200RM2U. Each unit individually provides enough capacity to power the entire rack. All of our new servers have redundant power supplies, so we will split each server's power feed over two UPSes. Besides making battery maintenance easier, this arrangement reduces the impact of single power supply, UPS, or circuit failure.

Here is a picture of the freshly installed units at the bottom of our server rack:

Although I really like the glass windows in my office, they do seem to make things chilly when it is cold outside.

The temperature when I walked into my office this morning was 59 degrees Fahrenheit. Space heaters to the rescue! This is a bit ironic considering I am located in a LEED certified building. I guess one way to stay green and minimize your energy costs is to eliminate the requirement to make the office habitable by humans.

Here is a pic of some of my tools for surviving in a wintery office environment:



Details about pictured items:

We received our shiny new servers from iXsystems (http://www.ixsystems.com). iXsystems "strives to provide the highest quality storage solutions, custom rackmount servers, pedestal servers, and appliances running FreeBSD, NetBSD, OpenBSD and various versions of Linux such as RedHat, SuSE, Slackware, and Debian." We went with a FreeBSD specialist vendor to eliminate the time required to track down hardware compatibility and to gain access to the reportedly excellent iXsystems technical support.

We ordered the 2U Jupiter server which is built on a SuperMicro platform. I wanted 8 drive spindles for storage system performance and 3.5-inch drives to keep overall price low. Some vendors have moved their 2U servers exclusively to 2.5-inch drives or their high performance SAS drives are obscenely expensive, so iXsystems met our specs and also came in a few thousand dollars cheaper per server than the larger vendors such as Dell or HP.



Initial casual benchmarking of the storage system and processor indicate that the machines are very fast. Disk i/o performance is fabulous. I have not yet found a load test I can throw at the machines to make them unresponsive.

Here are detailed specs: