So, yesterday I started working on a data server. I had one that we WERE using, but for various reasons I got frustrated and decided to “go all out” and buy some really expensive parts and piece together a more reasonable long-term data storage solution.
Original box: A8N-SLI Premium with an FX-55 and 4GB DDR-800 on Fedora 9 — also 650W PSU and a Thermaltake Xaser (I think it is a model ‘II’) case. I had the / fs on a software raid 1 between two PATA IDE drives (a 200GB and a 250GB … 200GB logical space). Then I had 5 SATA 750GB drives running software raid 5 (for a total of 3TB logical space). I also had a PATA 750GB that I was using for my /var/www/ mount point.
New Case
Story: We had a lightning storm take out the power at the apartment … multiple times. I got home from work one evening and the server was off. Not good. I booted it up .. and I noticed that my /dev/md2 (3TB raid5) mount did not mount correctly. I verified it existed correctly in /etc/fstab .. that looked good. I then proceeded to try to “recreate” its existence. Aparently, I did it incorrectly … long story short — I hosed my 3TB raid partition. So, I rebuilt it from “scratch” … with no data. I luckily had a backup of ~1/2 of the data (another amusing story). Moral of the story was … Per my previous notes, you notice that I make my original raid5 partition with 3 drives … then I grow it by adding 2 disks (1 at a time). I then show how to grow the xfs partition as well … all without unmounting. Its pretty slick really. However, one key thing that was lacking (originally) was the fact that we didn’t rewrite /etc/mdadm.conf. So, when the server rebooted, it was trying to initialize a 3-disk raid5 which no longer existed. I have since gone back and added the necessary changes to my guide, but it costs me a bit. I chalked it up to a learning experience. Moving on … I resynced all of my backup data to the newly formatted 3TB drive.
We have since moved to a house. The new house has some pretty serious power issues, which we readily found and tried to mitigate by using UPS’s. Strangely enough .. the same ones we had been using at the apartment. This was strike two … I assumed that the UPS was good and the system just shutdown improperly. We had fixed the software problem with the raid. Now … we had another power problem. The UPS worked (the first time). And then it didn’t. I woke up last Friday morning to a system that was powered off AND would not power back up. I assumed power supply, which I later confirmed.
So, with all of these problems … plus the angst of having to move all of my gear .. I decided to just “go all out” and buy an uber RAID controller and a sweet rack-mount case that could hold tons of sata drives. I did a bit of homework. I had been eyeing up an adaptec controller for a while. Here’s what I ended up getting:
Adaptec Raid Controller
Adaptec 51645 – Dual 1.2GHz cpu with 512MB DDR2 (yeah .. its JUST a PCI-E card). It occupies a PCI-E x8 slot and controls 16 internal and 4 external SAS or SATA ports via 4-to-1 cables that connect to mini-SAS ports. ~$825
SuperMicro Rackmount case with 15 SATA drives – This bad boy has room for 15 SATA drives and a triple-redundant, hot-swap 700W power supply. ~$750
So, I ordered both and awaited their arival. They were both supposed to arrive on Thursday. They did, but I ended up being sick that day … and I had them shipped to work, so I didn’t get to them until Friday. when I took a look I found a couple of things REALLY lacking. First, the Adaptec Raid controller came with NO cables … zip, zero, nada. I considered this to be REALLY lame considering the amount I spent for the thing. Somewhere along the line, I thought I read that it came with 2. Apparently I was mistaken. I took a gander around the web and came up with some $25 cables from PROVANTAGE. They should do the trick. I got 4 of them.
Second MAJOR thing was the front-panel cable on the server case is about 3-4″ too short. Entirely non-sensical. It was designed to work with THEIR branded motherboards. However, they sell the case separately without regard for a STANDARD installation location for a front-panel header. How stupid. At any rate, I found this page at Performance-PCS.com which has extensions for the 10-pin front-panel connector. By the look of it, its about 3-4″ of extension. I couldn’t be sure, so I ordered 2 of them. They were only $12.95+S/H each.
I was eager to get started. I thought … you know I could still install the hardware (Mobo/memory/CPU/expansion cards) and then install the OS and get all the drivers working right … even without all of those cables. First, I tried some modifications to my existing front-panel cable to make it longer. Lets just say, I didn’t have the appropriate tools and I just ended up smelling up the entire house of solder.
So I started putting together the parts. I purchased some DDR2 and I had an M2-based board laying around and I have a dual-core AM2 processor laying around … its like 2.3GHz. I was going to throw it together for this build. That did NOT end up happening. Couple of reasons, I didn’t remember that the board only had 2 memory slots. 2GB is probably enough, but the board isn’t designed for all of the function that I have in mind. So, I scratched that idea (plus I didn’t have a CPU fan/heatsink).
Bad CPU fan 
too tall.
Then I decided to verify that the “old” server JUST had a bad power supply. It was only the power supply … so I ended up deciding to just use it in the new case for now. After I secured it, I verified that it powere up properly. Then I went to connect the CPU fan when I noticed my next problem. It was too tall to fit in the case.
Talk about devastation and disgust. Sick days … down servers and poor part design have all added up to one fiasco of a project. The good news is, I have ordered 2 more CPU heatsink/fans and 3 more Power supplies and some thermal grease. So, when this is all said and done, we should have a few “extra” computers floating around for our LAN party days.
My goal for this machine: End up in my rack, hosting 3TB of logical space across a raid10 on 8 physical 750GB drives and running Fedora 9 hosting a couple of samba shares as well as some other random services that are key to our “base of operations”.
Patch panel and 48-port Gigabit (fully) switch
New Rack with 1of10 image processing servers
