From Documentation
Jump to: navigation, search
Note: This page's content only applies to the SHARCNET's legacy systems; it does not apply to Graham.


mako
Hostname: mako.sharcnet.ca
Target usage:
System information: see mako system page in web portal
System status: see mako status page
Real time system data: see Ganglia monitoring page
Full list of SHARCNET systems


System overview

This cluster is installed at the University of Guelph. The login node contains a dual core Intel Xeon CPU @2.33GHz with 2GB of RAM. Compute nodes mako[3-16] each consist of 2 quad core Intel Xeon Processors running @ 3.00GHz (8 cpus total per node) with 8GB of RAM. Compute nodes mako[17-32] each consist of 2 quad core Nehalem Processors running @ 2.1 GHz (8 cpus total per node) with 16 GB of RAM. All nodes in the cluster are connected via gigabit Ethernet. There is ~16TB of disk-space available on the cluster as a /scratch file-system, and it also mounts the SHARCNET-wide shared /home and global /work file-systems.

Support

Mako is not a production cluster and as such it is not supported at the same level as our other systems. Users should not expect immediate administration assistance to rectify problems with the system, for example, the Friday night before an assignment is due the following Monday. Being a training system, it is shared amongst several groups and prone to unintended abuse. As such, users should not expect a particular level of service or exclusive access to the system.

Any problems with mako should be reported via the SHARCNET web portal problem ticket system.

Accessing the System

This cluster can be reached on the SHARCNET network as mako or remotely as mako.sharcnet.ca

Upon logging into the system via ssh, users will find themselves on mako2. This is the login node for the cluster, and all work (editing files, compiling, submitting jobs) should be done on this node. Users should not have to ssh any further into the cluster.

Compiling

The Intel compiler collection is installed on the system. This includes C (icc), C++ (icpc) and Fortran (ifort). These compilers are installed in the user's default path.

To compile OpenMP code with these compilers, one should specify the –openmp compiler flag.

The OpenMPI MPI implementation is installed on the system to support MPI applications. To compile code with MPI support, one should use the MPI wrapper compilers that are found on the default path, ie. C (mpicc), C++ (mpiCC), F77 (mpif77) and F90 (mpif90).

Submitting Jobs

This cluster uses the standard SHARCNET scheduling wrapper SQ - please consult this link for usage examples and documentation. Users must submit all their work to the cluster with the SQ tools, it is inappropriate to run applications on the login node.

Scheduler Limits and Configuration

  1. All jobs on mako are limited to a 1-hour runtime limit
    • jobs will be killed automatically if they exceed this limit
  2. One should only submit OpenMP jobs with 8 or fewer threads since there are only 8 cores per node