From 94db48c0b70c2c84dcad7caf476c7a1cdf099d05 Mon Sep 17 00:00:00 2001 From: MB <michael.blaschek@univie.ac.at> Date: Tue, 18 Mar 2025 09:35:38 +0000 Subject: [PATCH] Update shebang in connect2jet script and enhance documentation on network drives and job queues --- SSH-VPN-VNC/connect2jet | 2 +- Servers/Aurora.md | 12 ++++++++++++ Servers/JET.md | 35 ++++++++++++++++++++++++++++++++++- VSC.md | 5 +++++ 4 files changed, 52 insertions(+), 2 deletions(-) diff --git a/SSH-VPN-VNC/connect2jet b/SSH-VPN-VNC/connect2jet index 49b4690..c69ce5f 100755 --- a/SSH-VPN-VNC/connect2jet +++ b/SSH-VPN-VNC/connect2jet @@ -1,4 +1,4 @@ -#!/bin/bash +#!/usr/bin/env bash # By Michael Blaschek # Date 18.01.2021 # CC BY 4.0 International diff --git a/Servers/Aurora.md b/Servers/Aurora.md index 90f44a0..bcb46cd 100644 --- a/Servers/Aurora.md +++ b/Servers/Aurora.md @@ -67,6 +67,18 @@ Paths: - `/srvfs/webdata` - `/srvfs/fastscratch/[username]` (SSD) +### Network drives + +Transfer of files between SRV and JET is **not** necessary. The file system is mounted on JET Nodes JET01/JET02 and vice versa. These mounted drives need to transfer the data via the network and latencies might be higher. Network connectivity between JET and Aurora can be 25Gbit/s. + +```txt title="Network transfer speed. 1TB of data." + | 1 5 10 15 min + 10Gb ----------------------------------- 13m20s + 25Gb -------------------- 5m20s +100Gb ------------ 1m20s +200Gb ---- 40s +``` +The 100Gb and 200Gb network connectivity is based on infiniband, which is even faster, for latencies. ## Software diff --git a/Servers/JET.md b/Servers/JET.md index 2a11830..74c01a5 100644 --- a/Servers/JET.md +++ b/Servers/JET.md @@ -71,6 +71,11 @@ Paths: - `/jetfs/shared-data` +???+ warning "JETFS on VSC" + Only from VSC5 you can access JETFS. Not the other way around. + + + ## Software The typcial installation of a intel-cluster has the INTEL Compiler suite (`intel-parallel-studio` or `intel-oneapi-compilers`) and the open source GNU Compilers installed. Based on these two different compilers (`intel`, `gnu`), there are usually two version of each scientific software. @@ -247,7 +252,7 @@ How to use a vnc server, go to [VNC](../SSH-VPN-VNC/VNC.md). ## Network drives -Transfer of files between SRV and JET is **not** necessary. The file system is mounted on JET Nodes JET01/JET02 and vice versa. These mounted drives need to transfer the data via the network and latencies might be higher. +Transfer of files between SRV and JET is **not** necessary. The file system is mounted on JET Nodes JET01/JET02 and vice versa. These mounted drives need to transfer the data via the network and latencies might be higher. Network connectivity between JET and Aurora can be 25Gbit/s. ```bash title="Mounted files systems" $ df -h @@ -260,6 +265,15 @@ $ df -h remjetfs 3.6P 1.6P 2.0P 44% /jetfs ``` +```txt title="Network transfer speed. 1TB of data." + | 1 5 10 15 min + 10Gb ----------------------------------- 13m20s + 25Gb -------------------- 5m20s +100Gb ------------ 1m20s +200Gb ---- 40s +``` +The 100Gb and 200Gb network connectivity is based on infiniband, which is even faster, for latencies. + ## Slurm The job manager is called slurm and is available on numerous other HPC systems in the EU. There are endless online documentations that can be asked for some guidance. Please have a look at the VSC tutorials or training courses. @@ -273,6 +287,25 @@ There is some more information about how to use slurm: - [Slurm Quick Start Guide - Manual Page](https://slurm.schedmd.com/quickstart.html) +### Queues + +We have several queue on JET that should allow optimal start for different scale jobs. + +| name | timelimit | nodes | exclusive | comment | +| ------- | --------- | ----- | --------- | ----------------------- | +| general | 3 days | 15 | - | **default** | +| devel | 30 min | 15 | - | higher priority | +| long | no | 16 | 1=jet03 | low priority | +| jhub | 2 days | 16 | 1=jet10 | reserved for Jupyterhub | + +There are two different node architectures, which can be used to stir computation on a specific microarchitecture. There are 7 intel (*skylake*) and 10 amd (*zen4*) architecture nodes available. One can select these specific architecture by providing additional parameters (constraint) to a job: + +```sh title="Slurm job contraint, microarchitecture" +#SBATCH --constraint="skylake" +#SBATCH --constraint="zen4" +``` + + ### Job efficiency reports since 2024 there is a new feature that allows to check how well one's jobs ran and get information on the efficiency of the resources used. The report is available once the job has finished. diff --git a/VSC.md b/VSC.md index 1905840..89e4b58 100644 --- a/VSC.md +++ b/VSC.md @@ -85,6 +85,11 @@ can be found on VSC-5 ``` +???+ warning "JETFS on VSC" + Only from VSC5 you can access JETFS. Not the other way around. + + + You can use these directories as well for direct writing. The performance is higher on VSC-5 storage. **This does not work on VSC-4.** ## Node Information VSC-4 -- GitLab