Presentation Supercomputing Zulip @ IAC SIEpedia IT News

Index of all issues November 2021 June 2021 March 2021 December 2020 Summer 2018 March / April 2018 January / February 2018 November / December 2017 September / October 2017 May / June 2017 January / February 2017 November / December 2016 July / August 2016
For earlier issues, please browse the Index

SIENews

The newsletter of the SIE de Investigación y Enseñanza N. 60 - November/December 2017

Python course for astronomy, February 26-28 and March 1, 2018

In the last few years the SIE has given several introductory Python courses focused on astronomy, after the ever increasing interest in this language by IAC researchers; as a matter of fact, last year we broke our own record by giving five Python workshops for astronomers, engineers and IT staff. Following this established tradition, we are organizing a new Python Course for Astronomy to be held on February 26, 27, 28 and March 1, 2018, in the Aula, from 10am to 12pm. This course is especially designed for users with little or no experience with Python (albeit some programming experience is required): after a broad introduction to the Python language, we'll focus on general scientific libraries and useful astronomical applications.
If you are interested in attending the course, just reply to the Forum thread below (you may need to create an user account if don't have one), where you will also find some aditional information: http://venus/SIE/forum/viewtopic.php?f=6&t=235

The new Severo Ochoa Supercomputer

A few weeks ago, a new Supercomputer, the "Severo Ochoa HPC", has been installed and is now fully operational. It comprises two nodes: a computing node with 192 Intel Xeon E7-4850 v4 2.10 GHz cores, 4.5 TB RAM and 40 TB disk space; and a login/GPU node with 20 Intel Xeon E5-2630 v4 2.20 GHz cores, 1 TB RAM and 11 TB disk space, plus a NVIDIA Tesla P100 GPU card. The most remarkable features of this new machine are the huge amount of memory available in the computing node (implemented as NUMA: Non Uniform Memory Access), shared among all cores — shared memory parallelism, such us OpenMP programs, can take good advantage of it —, and the GPU card.
Funds for this new machine were provided by three research lines of the Severo Ochoa Programme at the IAC (Cosmology and Astroparticles; Formation and Evolution of Galaxies; Solar Physics): while it can be used by any IAC's researcher, users belonging to the above research lines will be given higher priority. Detailed and exhaustive information about this new supercomputer is available at: http://venus/SIE/wiki/pmwiki.php?n=Supercomputing.Deimosdiva

LaPalma3 workshop

About 20 people from a dozen different Spanish institutions that belong to the RES (Red Española de Supercomputación) attended last week the three-day workshop about the setup and configuration of the new clusters, which are based on the former MareNostrum3 Supercomputer, after it was disassembled a few months ago. The workshop was held in the CALP (La Palma) by the BSC (Barcelona Supercomputing Center) staff, since LaPalma3 is the first of the RES nodes that has been already installed, and which will be in operation soon (likely in January 2018).
The new LaPalma3 Supercomputer consists of 252 computing nodes, with 16 cores and 32 GB RAM (2 GB/core) each. The total numbers are 4032 cores, 8 TB RAM and 346 TB disk space (implemented as a Lustre parallel file system) with a peak performance of 83.85 TFlops. This means that the new version of LaPalma is an order of magnitude more powerful than the old LaPalma (besides having 4 times as much RAM and 10 times as much disk space), with a bit lower total electric power consumption. Especially noteworthy is the fact that the new processors are Intel (specifically SandyBridge-EP E5-2670 2.6 GHz), so we hope that we'll have no more problems related to the old PowerPC CPUs used in LaPalma and LaPalma2 (big/little endian issues, lack of support, incompatible software, old compilers, etc.).
Once the LaPalma3 Supercomputer is running, we will give a talk to present it together with all other IAC's supercomputing resources that have been mentioned in this and past SIEnews.

Screen and nohup

Has it ever happened to you that you are running a very long program, something goes wrong with your terminal, and all the work is lost? Or have you run a program from the command line when using the VPN, then the VPN unexpectedly gets disconnected and your execution is gone? Usually this happens because your program is "attached" to the shell where you executed it, so logging out or killing the shell (like closing the terminal) will automatically terminate all processes that it started (programs receive the "hangup" (HUP) signal when the shell dies). There are several ways to avoid that:
  1. The easiest way is to prevent your program from receiving the "hangup" signal, so it will continue executing even after the parent shell is gone. You can simply place the nohup command before yours, and send it to the background with the & symbol: nohup ./your_program > output.log &
  2. Another easy way is to use the screen tool, which allows you to recover the current status of your execution later, as if you had never logged out or exited the shell. You can also create and name different sessions, display them in several windows, share sessions with other people, etc. For instance
    screen
    ./my_program
    Now your program will continue running even if you exit the shell, close the terminal or exit screen using Ctrl+A and d.
    To recover the session:
    screen -R
    If you had several active sessions, this last command will list all of them so that you can recover a specific one using screen -r <pid> See man screen for more options (i.e: -S to add a name to your session, -x for multi-display, etc.)
  3. There are some more tools for this task. For instance, tmux (Terminal MUltipleXer) has some more advanced features, like multiscreen with read-only permissions (that is very useful if you are teaching and you want your students see on their own screens what you are doing on your computer), etc.
SIE de Investigación y Enseñanza :: N. 60 - November/December 2017 - Contact: