Guillimin for Hackathon Participants

Print

Introduction to Guillimin for Hackathon Participants


The McGill High Performance Computing Centre has made some computing resources available for the Big Data Week 2015 hackathon. To get started using these resources, please read the following documentation carefully.


Resources available
Accounts and system access
Software and environment
Accessing the hackathon data sets
Setting up an iPython notebook
Accessing the GPUs
Using the Xeon Phi Co-processors
Contacting McGill HPC
Account and host assignment

Usage terms

These resources have been made available to hackathon participants by McGill HPC. By logging in and using these resources, you agree to the below terms.

  • These resources are to be used only for the purposes of completing projects related to the Big Data Week Montreal Hackathon.
  • Your usage will be monitored.
  • The resources will not be used for any illegal purposes.
  • Users will respect the privacy other users, and the underlying systems' data
  • Any violation of these policies will result in the immediate revoking of access for the hackathon team associated with the account and loss of any data not backed up off-site. There is no process for appealing our decision to revoke access.


Resources available

Each team will have exclusive access to a node during the hackathon with the following specifications:

  • Processors: Dual Intel Sandy Bridge EP E5-2670 (8-core, 2.6 GHz, 20MB Cache, 115W)

  • Memory: 64GB/node

  • Local scratch storage: 300GB (use the directory /localscratch)

  • Co-processors: 2x Intel Xeon Phi 5110P, 60 cores, 1.053 GHz, 30 MB cache, 8 GB memory, Peak SP FP: 2.0 TFlops, Peak DP FP: 1.0 TFlops


Additionally, we have made a smaller number of GPU nodes available via our scheduler system. These nodes each contain two Nvidia Tesla K20 GPUs, and otherwise have similar specifications to the exclusive nodes. Please see below for detailed instructions for accessing these resources.

For more information on Guillimin's storage options, please see our documentation.

 

Accounts and system access


Each team will be given credentials for an account by the event organizers. These accounts will allow ssh access to the node assigned to that team by the organizers. Note that ssh access is only allowed from a non-standard port.

$ ssh class01@aw-4r1[2-3]-nxx.hpc.mcgill.ca -p 57328


   
(see Appendix “Account and host assignment”)


(See Documentation on accessing resources)


Software and environment


The operating system on Guillimin  is CentOS 6.5 Linux. We have a large selection of software applications, compilers and scientific libraries installed which can be accessed by our module system. To see the full list, type ‘module av’ at the command prompt.


(See documentation on module use)

Note that multiple versions of some software are availalble. For example, to use python 2.7.3, you can use

$ module add python
$ python script.py

To use python 3.3.2, you can use

$ module add python/3.3.2
$ python3 script.py

Participants are welcome to install their own software tools. Installations must be done within the account’s home directory. Unfortunately, packages or installation managers that require root access cannot be used by hackathon participants.

Python modules can be installed using the ‘pip install --user’ command after loading the appropriate module (e.g. ‘module load python’).

$ module add python
$ pip install --user pandas

Similarly, the install.packages() function in R will notify the user of their permissions and ask to install in the home directory. The installation behaviour can also be controlled with the R_LIBS environment variable.

In general, please consult the installation instructions for the relevant software packages for installation options and details on how to install them with normal user privileges.

Please contact us if you have questions about installing software packages.


Accessing the hackathon data sets

The data sets for the hackathon are accessible through the following directory:

/gs/project/bigdataweek

The data is stored on a high-performance, parallel file system called GPFS.

All participant accounts have read-only access to these data directories. Analysis scripts can directly reference these directories, but all scripts and output must be stored in the account’s home directory.

Other standard filesystems

Filesystem Size Description
/home/class## 10GB Home directory
/localscratch 300GB Local hard disk
/dev/shm 32GB RAM disk


Important: Please make regular offsite backups of your work during the hackathon. Please retrieve any data you wish to preserve before the end of the hackathon.

Setting up an iPython notebook

1. ssh to the assigned node
2. run the script notebook.sh from the home directory. This will start the IPython notebook on port 8088 with a self-signed certificate. You will be prompted to select Python 2 or 3
3. access SSH via: https://node-name.hpc.mcgill.ca:8088 and accept the certificate. Note that Safari will not work with a self-signed certificate. Firefox and Chrome do allow that.
4. Enter your class account password to authenticate to the notebook
5. Leave the ssh window open. If the notebook server crashed, just rerun the script

Accessing the GPUs

Nodes containing 2 Nvidia Tesla K20 GPUs can be accessed for periods of up to 2 hours at a time through our scheduler. You may get an interactive shell on the GPU node for two hours (or until the session ends) by executing the following script in your home directory:

# ./gpu_access.sh

Note that depending on the demand, there may be a waiting period for GPU resources. Please submit only one GPU job per team at a time. Wait for your job to finish before submitting a new one.

(See our full documentation on using GPUs on guillimin)


Using the Xeon Phi Co-processors

The exclusive node assigned to your team contains two Xeon Phi co-processors that can be used to accelerate your applications. Please see our documentation for instructions, and our training materials for a tutorial.


Contacting McGill HPC

Participants are welcome to ask questions and get support by emailing This email address is being protected from spambots. You need JavaScript enabled to view it. . Additionally, McGill HPC will have staff members intermittently available on-site during the hackathon to help users access the data and resources and to answer any questions.

Appendix A - Account and host assignment

Username

*.hpc.mcgill.ca

Team

class01

aw-4r12-n01

 Organizers

class02

aw-4r12-n02

 

class03

aw-4r12-n03

 

class04

aw-4r12-n04

 

class05

aw-4r12-n05

 

class06

aw-4r12-n06

 

class07

aw-4r12-n07

 

class08

aw-4r12-n08

 

class09

aw-4r12-n09

 

class10

aw-4r12-n10

 

class11

aw-4r12-n11

 

class12

aw-4r12-n12

 

class13

aw-4r12-n13

 

class14

aw-4r12-n14

 

class15

aw-4r12-n15

 

class16

aw-4r12-n16

 

class17

aw-4r12-n17

 

class18

aw-4r12-n18

 

class19

aw-4r12-n19

 

class20

aw-4r12-n20

 

class21

aw-4r12-n21

 

class22

aw-4r12-n22

 

class23

aw-4r12-n23

 

class24

aw-4r12-n24

 

class25

aw-4r12-n25

 

class26

aw-4r12-n26

 

class27

aw-4r12-n27

 

class28

aw-4r12-n28

 

class29

aw-4r12-n29

 

class30

aw-4r12-n30

 

class31

aw-4r13-n01

 

class32

aw-4r13-n02

 

class33

aw-4r13-n03

 

class34

aw-4r13-n04

 

class35

aw-4r13-n05

 

class36

aw-4r13-n06

 

class37

aw-4r13-n07

 

class38

aw-4r13-n08

 

class39

aw-4r13-n09

 

class40

aw-4r13-n10