Your First Cluster: Build or Buy?

July 20th, 2009 8:35 am
Posted by Douglas Eadline
Tags: , , , , , ,

Consider the time -- as well as the cost the cost -- when building your own cluster

Many people exploring cluster computing often consider building their own system as a means to reduce costs. When asked about this idea, I offer the following opinion, If you want to learn about how to build a cluster, then build your own. If research and engineering are your goals, then it is best to buy a pre-configured cluster because you are unlikely to save any money and you will be forced to learn about things and apply time/money to issues that are not directly related to your goals. I emphasize that building your own hardware is worthwhile educational experience in its own right, but keep in mind the cost savings often come with additional time and effort on your part.

Some people are perplexed by my answer and I explain that the "cost" to build your own cluster is much more than that of a pile of cheap parts from the lowest bidder. First, there is the time needed to design the hardware. Next, comes the actually construction time (provided all the parts show up), and finally there is the time needed to make sure everything works (after initial failures and exchanges). In terms of calendar time, this can be as long as six months. And, even if you have low cost labor (i.e. students) there is always the delay with your primary research objectives.

If you want to learn about computer BIOS, motherboards, memory, power supplies, part failures, drivers, power and cooling, and RMA policies, by all means build your own nodes. At first it may seem like a interesting and fascinating process, but after the fourth or fifth system it may become a bit tedious and the value of preconfigured and assembled systems will become obvious. Of course, you can always buy nodes pre-built, but be careful. HPC usually pushes servers more than other applications. In addition, it is important to make sure the hardware is identical down to the motherboard and BIOS revisions. Slight variations could be the root of future problems. Finally, make sure your vendor guarantees that all the components will work together (i.e. nodes, switches, cables, cable lengths, storage systems, etc.), provides an adequate service/replacement policy, and will work with you in the event of problems.

If you need to start folding proteins or designing airfoils right away, then your best bet is to get a pre-configured cluster that is ready to run the applications you need. One of the best ways is to look for a vendor that sells Intel Cluster Ready systems. These systems are turn-key and ready to run. Of course, you can always use freely available software packages and cluster tool kits for any pre-built cluster. And, remember when it comes to the build vs buy decision it more a question of time than of money.

Comments

Comment from zavvyone
Time July 22, 2009 at 11:17 am

SGI is an Intel Cluster Ready vendor--learn more at: http://www.sgi.com/products/servers/altix/icr/ or email me if you need help from some of the best in Linux Cluster engineering!

Pingback from Cluster Connection » Your First Cluster: Build or Buy? | Linux Affinity
Time July 27, 2009 at 6:27 am

[...] is the original post: Cluster Connection » Your First Cluster: Build or Buy? Posted in: Server, Super [...]

JOIN THE CONVERSATION


You must be a Registered Member in order to comment on Cluster Connection posts.

Members enjoy the ability to take an active role in the conversations that are shaping the HPC community. Members can participate in forum discussions and post comments to a wide range of HPC-related topics. Share your challenges, insights and ideas right now.

Login     Register Now


Author Info


Dr. Douglas Eadline has worked with parallel computers since 1988 (anyone remember the Inmos Transputer?). After co-authoring the original Beowulf How-To, he continued to write extensively about Linux HPC Clustering and parallel software issues. Much of Doug's early experience has been in software tools and and application performance. He has been building and using Linux clusters since 1995. Doug holds a Ph.D. in Chemistry from Lehigh University.