Flux

Flux is a collaboration between the AVP-RCI, ITS, the Center for Advanced Computing at the College of Engineering, LSA, and the Medical School to provide a high quality High Performance Computing cluster environment for researchers at the University of Michigan Ann Arbor campus and their collaborators.

Flux (phase) II is currently available.  Flux II is an interim solution that allowed us to verify the fully built out system's design and cost model. Flux II is currently scheduled to be phased out and replaced by Flux (phase) III in the Fall of 2011.

Flux (phase) III will be the fully built out version of Flux.  Its rate will eventually fully reflect its costs, with gradual rate increases planned to start in July 2012.

Access To Flux 

Primarily, access to Flux requires a purchase of an allocation, with a unit of allocation being 1 core for 1 month.  The rate mentioned below is for an allocation of resources and time and is not based on actual usage.  The charges will appear on your monthly Statement of Activity after every month in which you had an allocation.   To inquire about flux allocations please email: flux-support@umich.edu.

Secondarily, you can use the Flux On-Demand service, which allows you to pay for what you use.  This allows accounts with bursty usage patterns to use Flux and only pay for actual usage.  However, there are some limitations to the system:

  • Limits to the total number of processors collectively used (currently 50)
  • Limits to the number of processors used by any one user (currently 25)
  • Jobs run in the On-Demand service are scheduled with a lower priority than jobs using the allocation system.  Once the jobs are running, they will run identically to an allocation-based job.

Accounts that have workloads that require predictable start times should look to purchase an allocation.  On-Demand usage for the month will be rounded to the nearest tenth of a core/month.  The resulting charges will appear on your monthly Statement of Activity after every month in which you used the On-Demand service.   To inquire about flux allocations please email: flux-support@umich.edu.

The University approved rate for Flux II is $11.20/core/month.  For your convenience the rate can be represented as a hardware component at $6.81/core/month and the facilities component at $4.39/core/month. 

Advantages of using Flux

  • Since you're buying an allocation, you can purchase exactly what you need for the length of time you need it, including short-term increases as your requirements change. 
  • Flux is a large pool of computers; if a machine goes down, your allocation will be unaffected.  The scheduler will just find another machine for your jobs to run on. 
  • Allocations are cumulative, so you can have multiple funding sources.

Flux Training and Support

Training is available for all users of Flux. Training inquires and support requests should be emailed to: flux-support@umich.edu.

Flux II Hardware Highlights

  • Shares head node, queuing, and software infrastructure with the CAC cluster Nyx
  • Dedicated login and transfer hosts (flux-login.engin.umich.edu and flux-xfer1.engin.umich.edu).
  • Dual socket six core Intel Core I7 CPU nodes providing 12 total cores per node
  • 48GB of RAM available on each 12 core node
  • 40Gbps Infiniband networking between all nodes for fast MPI communication
  • Expanded Lustre storage infrastructure to the CAC's /nobackup file system for a total of 143TB

Software on Flux II

Flux II allocations share the same software as the CAC cluster Nyx. Software is accessible via the module system. Some units also provide their own software not managed by the CAC. This requires that unit's module to be loaded before the units software is available. An example is loading the lsa module. The module will add, in addition to the stock CAC modules, any software that the LSA stewards of Flux have added.

Queuing on Flux II

Flux II is accessed via the Nyx login nodes and is shared by the CAC cluster Nyx. Flux II has its own queue called flux which should be used in your pbs file.  Also all Flux users must have an account with valid allocation. For more details on queuing options see the Nyx PBS documentation.

Once you get your flux allocation (in this case, let's call it "example_flux"), there are three things you'll need to change in your PBS script to use flux - the queue, the account, and the qos:

#PBS -A example_flux
#PBS -l qos=example_flux
#PBS -q flux

  

Planned Changes for Flux III

As we expand Flux to its self-sufficient state, we are planning on increasing the rate over the course of 2-4 years to reflect to full costs of operating a high-performance computing cluster.

Flux III will also have its own software library, independant of that maintained by the College of Engineering on Nyx.  We expect this to be quite similar to Nyx's software library, perhaps with more focus on a broader selection of software.

Flux III will have its own storage, networking, login and administrative hosts and will not share that with the College of Engineering's cluster, Nyx.  This will allow for a more transparent cost structure and more independance for Flux than is currently possible.