Department of Electrical and Computer Engineering
Information Infrastructure Institute

Guidelines

CyBlue is the new IBM Blue Gene supercomputer at Iowa State University. It is meant for enabling cutting edge research at the frontiers of science and technology. As the most powerful computing platform on campus, it is intended for computations that cannot be carried out on any other machine on campus. This will be the guiding principle in granting account requests.

It is envisioned that CyBlue will be used for a limited number of high profile, large-scale projects that will define the best of high performance supercomputing research at Iowa State University. It is important that work that can be carried out on small clusters should be carried out on them, even for those who have access to CyBlue for another large-scale project. The only reason small partitions of CyBlue are sometimes made available is to support application porting and to carry out BlueGene/L specific optimizations. It is strongly recommended that initial code development be carried out on clusters and other platforms.

When is it appropriate to request an account on CyBlue?

To request an account on CyBlue, you are required to submit a brief justification. Some legitimate needs are outlined below:

1. You are working on a problem whose scale can only be tamed with CyBlue.
2. Your work will push the frontiers of supercomputing applications in your field.
3. You are preparing to enter a competition that will bring significant visibility to your group and Iowa State. Ex: Gordon Bell award, CASP etc.
4. You are an application developer and wish to demonstrate scalability on massively parallel platforms.

When is it not appropriate to request an account on CyBlue?

CyBlue is not meant for small problems and applications with low scalability that are best conducted on small or moderate sized clusters. Here are some examples of when not to request an account on CyBlue.

1. You like to get your hands on every multiprocessor system and run some serial jobs.
2. Your job can be run on a cluster but it is good to have accounts on multiple systems to increase chances of having processors available.
3. You never intend to use more than 128 processors.
4. Your code does not scale beyond 128 processors.
5. Your needs can be met by a cluster but you don’t have one and no one is giving you access.
6. Your code is not MPI-based (only MPI is supported on CyBlue).
7. Your idea of parallel processing is to have all data available at every processor and just partition the computations (CyBlue only gives you 512 MB/node that you will share with the OS).