1250days since
SC10 SCC Competition


Application specific questions can be found on the pages associated with the specific applications.  This FAQ will cover all other subject areas of the competition.

  • For the competition, we have a 26 amp limit.  We also know that we will be monitored by 2 PDU's with 13 amp limits.  Do we have to stay under 26 amps total, or do we have to stay under 13 amps on each PDU?
    You man not draw over 13 amps on either PDU.You may draw 13 and 13,  you may draw 10 and 12, you may draw 13 and 12, you may NOT go over 13 on either PDU. i.e. 14 and 10 is out. as is 14 and 13, etc. etc.

  • What is the space setup?
    The exhibit hall floor plan is attached below in the attachments.  We are located on the far left wall.  Each team will be given a 10 x 10 booth space to contain their machines and students.  Each booth will be furnished with two tables and 6 chairs.  There will also be a control room and an additional communal seating area.

  • Do you have a recommended scheduler?
    There are open source schedulers available for teams to use (i.e. OpenPBS, Maui, Torque, etc.) along with software provided with various linux distributions.  We do not have recommended vendors.        

  • If we're trying to conserve as much power and maximize processing power for the competition applications itself, how do we do visualization at the same time?

    Teams will be provided with three 20 Amp 120V circuits (each with a soft limit of 13 Amps) on separate breakers.  Two will be used for the competition cluster and cannot exceed the 26 Amp limit.  The third may be used for the LCD, laptops, coffee maker, etc. We will provide two PDUs for the competition cluster to monitor the power usage.  Any power strips and additional visualization computers needs to be supplied by the teams.

  • How have teams done visualizations in the past?  Have they had real-time visualizations using their cluster, or computed everything ahead of time and had stuff playing on a video?

    Last year, VISIT was one of the competition applications, so they had to use it in parallel to create visualizations. This year we do not have a visualization application, but the application runs provide output that can most certainly be visualized.  What you decide to do will be part of your interview&visualization score.  You can create canned images to run on the LCD, but you are most certainly welcome to learn the visualization software and run those as well.

  • Is there a restriction to using a Drobo unit to move our files from the cluster to an off cluster archive unit? We are assuming that this would not be included in our power restrictions for the cluster correct?

    "It depends what you are using it for and how".

    If the drobo unit is being used to purposes of running their cluster -  i.e. backup, imaging, or if it being used as  storage which is accessed and used by computations (i.e. intermediate files, dictionaries, point-in-time snapshots, output of results, etc.) , it needs to be plugged in and counted for power with the cluster.  If this is an archive used for visualization purposes, then it  may be plugged into stuff along with visualization laptops.

    For example - they should not be storing computation dictionaries (i.e. for password cracking) on storage off the cluster on this unit and accessing them directly.  You should not be using this to image your cluster machines.

     Sometimes this is a bit of a gray area - but try to keep within the spirit of the rules which is "ALL YOUR COMPUTATION IS DONE WITHIN THE POWER BUDGET" - the external power is for your laptops and making eye candy to impress people.  If the drobo unit was picked up and taken away, and all you would lose would be eye candy - then it is probably just fine to have powered externally.  On the flip side, if the drobo unit dissappeared mid competition and you would NOT get as much computation done as you otherwise would - then likely that unit should be powered with the cluster - as you are using it for storage needed by your computations. 

    So our official answer is "if in doubt, power it with your cluster to avoid dissappointment".

  • What plug should we use for the system? Maybe L5-20? Or is it going to be a normal 3-point plug?

    Your system will be monitored using the following PDU  http://www.apc.com/resource/include/techspec_index.cfm?base_sku=AP7801 .  Your plug should be consistent with it.

  • Is it OK if we control the power on/off of part of the system via software in order to manage the power limit? Or do we have to go full-on from the beginning? 

    It has to be "on" and in the computer for the power-monitoring phase to be used afterward.  For example, if you can power 12 nodes, and run HPL on 11 of them, leaving 1 idle, then it's fair game to use. (Purdue did this in 09).  You you can use things like turning down the CPU speed via /proc to lower power usage, but not (for example) ipmitool to power off/on, or removing RAM for the HPL and phase and putting it back in after power is no longer being monitored.

  • How much disk space should we have to store all of the model output?

    In the past a few hundred gigabytes was sufficient, but as problems become more complicated and technology improves the demand also increases.  We highly recommend teams study the applications and test problem sets found (given by us and found online) to determine an appropriate disk space capacity.

  • Can students be on the floor 24/7? or should they leave the floor in the night and come back next morning?

    Teams will have 24/7 access to the SCC competition space starting Saturday (11/13) to Thursday  (11/18).  We expect advisors and student teams to monitor their time on the floor to ensure maximum safety.  Team members should take shifts so that a single student is not awake for longer than is safe.

Hai Ah Nam,
Sep 10, 2010, 6:45 AM