Turing: Class structure - interactive and batch limits

*****************************************************************************

Structure of classes on Turing				   15th of July, 2014

*****************************************************************************

I.   Blue Gene/Q classes
     ===================

  T(h)
      ^
      |
  20h +    +---+--------+---------+---------+---------+
      |    |   |        |         |         |         |
      |    | S |  MRt3  |  1Rt3   |  2Rt3   |  4Rt3   |
      |    | B |        |         |         |         |
  10h +    | M +--------+---------+---------+---------+
      |    | R |        |         |         |         |
      |    | t |  MRt2  |  1Rt2   |  2Rt2   |  4Rt2   |
      |    | 3 |        |         |         |         |
   1h +    | 2 +--------+---------+---------+---------+
      |    | 1 |        |         |         |         |
  30m +----+ 0 |  MRt1  |  1Rt1   |  2Rt1   |  4Rt1   |
      |        |        |         |         |         |
    0 +----+---+--------+---------+---------+---------+
      1    16  64       512       1024      2048      4096
                                   Number of compute nodes
Core ->  base unit
Compute node ->  16 cores
MR ->  Midplane:  512 nodes (= 8192 cores)
1R -> 1 rack of 1024 nodes (= 16384 cores)
..
T(h):  Elapsed time in hours

Memory:  16GiB per node (1GiB per core)

Please note:
------------
- The number of compute nodes reserved must be 1, 2, 4, 8, 16, 32, 64, 128,
  256, 512, 1024, 2048 or 4096.
- If less than 64 compute nodes are reserved, network resources (5D-torus)
  could be shared with other jobs. This could have an impact on the MPI
  communication performances and on the I/O (1 I/O node for 64 compute
  nodes).
- If less than 16 compute nodes are reserved, the maximum elapsed time is
  limited to 30 minutes.
- Caution: In the accounting of hours used, the Elapsed time is multiplied
  by the number of reserved cores even if some of them have not been used.


II.  The front-end (Power7 p720) classes:
     ------------------------------------

There are 5 classes available for your uniprocessor activities such as batch
cross-compilation, file transfer, small sequential pre- and post- processing
in relation to your massively parallel jobs:

- t1 and t1L: limited to 1 hour of elapsed and CPU time.
- t2 and t2L: limited to 10 hours of elapsed and CPU time.
- archive:  only for (mfput/mfget) file transfers with Ergon.

The time spent in the front-end classes is not accounted for.
Memory by default (stack and data): 3.5 GB. For the classes t1L and t2L, it is
possible to request up to 10 GB via the keyword:
                         # @ as_limit = 5.0gb 

The class “archive”:
This class is STRICTLY RESERVED for (mfput/mfget) file transfers with Ergon.
IDRIS can suspend running the jobs in this class when the Ergon machine is not
available.  When the machine is re-started, the jobs will resume being run in
the usual way. In order to use this class, it is necessary to specify the
following LoadLeveler keyword in your job:
                         #@class = archive 

III.  Interactive limits on Blue Gene/Q
      =================================

We do not offer the possibility of running executable files in an interactive
way on the Blue Gene/Q machine.

IV. Interactive limits on the front-end (Power7 p720)
    =================================================

Memory (stack and data): 3.5GiB

V. Bonus jobs
   ==========

It is possible to run so-called "bonus jobs" by adding the LoadLeveler 
keyword (before #@queue):

  # @ account_no = bonus

Bonus jobs are not included in the DARI hours allocation and
accounting.  They are run during machine low-load periods. Bonus jobs
are limited to a maximum of 512 compute nodes and a maximum elapsed
time of 20 hours (see www.idris.fr : Turing -> Bonus job)


To see the French version, type : news class
*****************************************************************************