The pride and joy of my computational accomplishments is Sherlock. Sherlock has always been a cluster of computers meant for the purpose of heavy number crunching. If you are reading this, then Sherlock is online.
The original version of Sherlock was a series of ten Dell Optiplex 320s stacked on top of each other. This primitive beowulf cluster was based on the wonderful Microwulf project. I had a lot of fun with building that system, but truth be told, I never got it working as well as I would have liked. Being an eighth grader at the time, I was way out of my league with a non-existent budget. I would later find out that most of problems came from a faulty SCSI controller connected to the data host, but I had already given up on the project, so it was a moot point. I wish I had pictures of the original Sherlock, but I did not have a camera at the time. I still have the motherboards from the project, completely stripped of the Dell chassis, waiting to be reused.
The Second, Infamous Edition
The current version of Sherlock makes me incredibly happy at how it is working. After I got to Purdue I wanted to rebuild a cluster system for data crunching. I decided upon retrying the Sherlock concept in a dorm room, which was an incredibly ignorant idea. I am still grateful to my Freshman roommate for not killing me, as this hunk of junk was large, hot, and noisy. Purdue has what is called the Surplus Store, and most of the computers I used in the second iteration was from here. Thus the reason some of it still has “Property of Purdue University” stickers on them. The main components of this revision were two Sunfire X4600 M2s, two BlueArc 160s, one NetApp DS14 MK4, with fourteen 450GB fibre channel hard drives. The two Sunfires were the main computing part of the system, each having eight dual-core Opteron CPUs with 8GB of DDR2 RAM per CPU. I liked those server a great deal, as they run like they were on fire, and to be honest, it would not have surprised me if they did actually spontaneously combust. With four 1000W power supplies, I was not exactly saving the earth with these things, but at 50$ a piece from the Surplus Store, I could not beat the price for a broke college kid.
The Third, Product of a Calmer Time
Being that the second version sounded like two Boeing 747s fighting for air supremacy directly above a fireworks festival, my roommate kindly suggested it was time for a change. I was happy to oblige, as keeping the window open in the middle of Indiana weather left me questioning my sanity. I got a job at Wiley Dining Court and saved every penny I earned for four HP Proliants, a Dell PowerConnect, and a heap of hard drives. I ended up working more hours than I thought humanly possible (while also having 18 credit hours) and bought everything I wanted. This is my current system, as I bought completely overkill hardware for the interests I have. The HP Proliants each came fully stocked with two 3.6GHz quad-core CPUs, 72GB DDR3 RAM each, and four gigabit NICs. I also ended up getting a Dell PowerConnect 6248, and a Supermicro SAS backplane with seven 3TB Seagate SAS drives. I could not be more happy with the system I currently have and have built from scratch.
Sherlock Software and Implementation
The hardware would be absolutely useless without a purpose, something hard to define with this system. My father always used to ask “But what does it do?” For the first system, I had no answer. It was cool, I wanted to build it. That’s about it. If I remember correctly, the most productive system I got to run on the first cluster was IIS (Windows Web Server) and an SMTP (E-Mail) server. That was it. In my defense, the fastest internet speed the house had at the time was 56Kbps dial-up. (990Mbps hit me like a brick at Purdue) So much for the power bill…
The second system was a bit better. I had just figured out the wonders of high-level scripting languages like Python and MATLAB, and I was enamored with the possibilities of Neural Networks and Expert Systems, so all the systems were running Ubuntu 16.04LTS. I never ended up having enough time and effort to put into the systems, so they never ended up doing that much.
The current system is actually useful. To understand what is going on, I had to hand-draw a diagram, but after all, a picture is worth 1000 words.
In terms of what it does for me currently, here is the short list:
- Current Desktop
- Home Theater (Plex)
- Always on Backup
- Secure Network Attached Storage
- Virtual Private Network (OpenVPN)
- Web Services (Apache, Nginx, Postfix)
- P2P Services (Dtella, Bittorrent)
- And Room Heater
To be honest, its strangely nice to have my one section of the network where I know everything is going to run just the way I want it too. The frustration saved from trying to use Purdue’s Software Remote systems is enough to justify owning an entire cluster for just the tiny space of a dorm room.
As of June 2019, Sherlock Mk3 was retired in favor of Laplace’s Demon, a smaller system with significantly more compute power. I still have all of the components of Sherlock, and am grateful for all the things it taught me over its 8 years of existence. Reality is often disappointing, but having 700lbs of computer equipment while being a working road warrior just clashes too hard.