The “budget” Matterhorn server: Zoidberg

Opencast Matterhorn uses a server or cluster of servers for media transcoding, storage, and distribution. For the internal pilot of Matterhorn we’ve decided to go with a single server instead of multiple servers in a cluster. The advantage of using a cluster is scalability to accommodate more media and viewers in the future. The advantage of a single server is lower initial cost and simplified configuration. Demand might necessitate moving to a cluster down the road but in the foreseeable future it should cover our pilot needs. After the pilot period the server infrastructure will be evaluated again with more experience in using Matterhorn to help guide decisions. During the pilot period we expect approximately one hundred students to utilize the system and two or three classrooms to be captured. Larger organizations should probably consider a cluster configuration to keep up with the workloads.

Primarily for flexibility and cost effectiveness we built our own server from parts starting with a Super Micro chassis. Parts were sourced from Newegg and Dell based on pricing and availability but the components should be available from most computer part distributors. The specifications of the system are 16 processor cores at 2.0 GHz, 16GB of DDR3 registered memory, 6TB of usable RAID 6 storage, and expandable to 12TB of usable RAID 6 storage. The total cost of the server was $4,161.55 including shipping for the parts plus an afternoon to assemble and install. Considering the performance and storage capacity for the investment, we feel its a pretty good deal and well suited for hosting a small to medium sized Matterhorn installation. A list of components is below linked to the manufacturer’s site plus the price in we paid in July of 2010 (without shipping).

Super Micro has an exceptional track record for reliability and support. All of the cables (even for the drive bays) are routed through the chassis and plugged in which makes setting up the server straightforward and quick even for those with little server hardware experience. That particular server also includes a redundant power supply and several PCI-Express slots for RAID controllers. Essentials like the processor heatsinks, server mounting rails, and DVD drive come with the server. Opteron 6100 processors were selected based on the performance for the physical size and cost, they are about as dense and inexpensive as it gets in the server market at the time the server was purchased (up to 12 cores per processor). The Adaptec RAID controller was selected because the mainline Linux kernel has drivers for most Adaptec cards natively (also most 3Ware and LSI controllers), the SAS connector matches the cables in the chassis, and if necessary later the controller can be installed in a low profile expansion slot system. Seagate Barracuda XT drives seem to be a good balance of performance and reliability to cost. Traditional server drives with SAS and 10-15K RPM get expensive very quickly and don’t add a lot of value to the server for our purposes. Last but not least the Kingston memory was chosen because it was very inexpensive and it works. Depending on the supplier, there might be other memory modules that are less expensive and do the same thing.

Ubuntu 10.04 LTS will be used for the pilot and likely for the production Matterhorn installation. Why Ubuntu? There’s a great community surrounding Ubuntu with many helpful users, the Matterhorn development team has thus far provided instructions that work with Ubuntu for installation and configuration, and its an LTS (long term support) release which means security updates will be available until 2015 (five years). Other freely available distribution like Fedora and the non-LTS releases of Ubuntu have fairly short life cycles which can make it difficult to maintain later down the road. CentOS was also considered for an operating system but since RHEL/CentOS 6 is just around the corner the support lifespan won’t be as long as Ubuntu 10.04 LTS. More specifics on the software and installation will be covered in future posts.

A few other notes to add to the mix. One of the disks is a hot spare so the RAID can rebuild itself automatically if one of the array disks fails, otherwise the usable storage on the server would be 8TB instead of 6TB. RAID 6 was used for various reasons, whether or not its necessary we’re not sure yet. Though we’d rather error on the side of caution than spend a week recovering data from tapes and losing a week of captured lectures. We called the server Zoidberg because we love Futurama. As progress is made on implementing Opencast Matterhorn here we’ll keep the community informed as well as keep a running total of expenses so others can get an idea of what it takes to implement a similar setup.

Permanent link to this article: http://sites.tamuc.edu/innovations/the-budget-matterhorn-server-zoidberg/

Leave a Reply

Your email address will not be published.