MFCF Grad Session 2017

AGENDA

• Introduction - 4th Annual Grad Computing session - Steve Vavasis, Associate Dean, Computing

• Help Centre and requests

• Department Grad reps

• Graduate student computing

• desktop, software, printing, and general help, web pages

Available resources in Math

• Data storage - how to safely back up your data

• Math Windows environment

• Microsoft Remote Desktop to access Windows - including “Mondo”

• Remote applications

• Math Linux environment

• GPU computing, specialty Linux machines, High Performance Computing (HPC)

• Questions ??

MFCF HELP CENTRE• MC 3017 x36323

• Help with desktops - reimaging, admin access (not Stats & ActscSci), download software

• Laptop connections - wired connections

• Accounts, Printing, Nexus passwords…

• Grad rep - meetings once a term

• communication between MFCF and grads regarding computing - new developments

• Request Tracker

• Lynda.com - online training

RESEARCH WEB PAGES

• encourage web pages to be built on IST Scholar for university wide common look and feel

• https://uwaterloo.ca/scholar/

• does not require programming or coding skills

• easy to use tools, self managed with templates/content modules

• bio, CV, publications, events, etc….

• publications can be imported into UWaterloo Scholar

• Documentation and guidelines:

• https://uwaterloo.ca/web-resources/scholar#documentation

DESKTOP - more than 400 Mac minis deployed - NOT BACKED UP, I repeat NOT BACKED UP

HOW TO ACCESS YOUR RESEARCH STORAGE

• Use Virtual Private Network (VPN) when accessing all campus resources from off campus - https://vpn.uwaterloo.ca

• storing files on Math file server (backing up your data):

• Mac mini > files.math icon • Mac standalone: Go > Connect to Server….

smb://files.math.uwaterloo.ca/<UWuserid>

• Windows Remote Desktop: windows.math.uwaterloo.ca P: drive (files are stored automatically in your file space on the fileserver)

• Windows standalone: Map a network drive \\files.math.uwaterloo.ca\<UWuserid>

• Linux: File Manager > Connect to Server…. smb://files.math.uwaterloo.ca/<UWuserid> • ssh to linux.math

ALTERNATE DATA STORAGE

• O365 vs Office 2016 • OneDrive vs OneDrive for Business (UW vs Canada vs Other)

• https://portal.office.com         (Nexus credentials) • https://mysharepoint.uwaterloo.ca

• convenience vs security • web interface vs synchronized data directories • offline drives / storage caches • conflicting storage methods

• Message from AD Computing, Information Management (Policy 46)

by: Jim Johnston

WINDOWS ENVIRONMENT

Brief overview of Microsoft Windows Services

Name (.uwaterloo.ca) Purpose auto-logout Comments windows.math Research computing cluster no Mac mini icon

windows.student.math Student/course computing after 2 hrs No research disconnected computing.

mondo.math Research GPU computing no Reservation only. GPU-centric software.

remoteapp.math Research computing integrated after 30 mins More RAM, CPU, into other environments no open apps local temp storage

by: Jim Johnston

REMOTE DESKTOP

All regions:

• must use campus VPN to access from off-campus • use Nexus credentials • require Microsoft Remote Desktop Client (version >= 8); rdesktop on Linux • minimum installed software:

• MS Office 2016, Maple, MATLAB (requires license through IST web store), R/R-studio, Anaconda (Python, Jupyter), etc.

• automated network drives • P: files.math - Math Research • M: mfsfs - Math students/courses

• OS Patching requires reboots • save your work often, don’t leave files open unnecessarily • write your code with checkpoint methods • don’t ignore email about reboots

MONDO

What is mondo?: • Research GPU Windows computing

• Windows 2016 PC: 32GB RAM • Intel Core i7-7700K CPU @4.20Ghz • 240GB PCIe SSD • 2TB SATA HDD • 8GB NVidia GeForce GTX 1070 (Pascal GPU, 1920 CUDA cores)

What is mondo used for?: • test code before sending to linux.math or SHARCNET • test code that needs graphics console • test code that is highly disk I/O intensive

REMOTE APPS

Remote Application Servers (remoteapp.math): • newest software deployed here • more RAM, CPU, local temporary storage • one session accessible from many devices • disconnect from phone/tablet to reduce data usage fees • all access methods are equivalent and require Microsoft Remote Desktop

Client (version >= 8)

REMOTE APPS

• Three ways to connect •Web page

• h"ps://remoteapp.math.uwaterloo.ca/ •Microsoft Remote Desktop Client (RDC) • All platforms except Windows use "Remote Resources" menus

• Integrated into Microsoft Windows • Control Panel -> "Search" for “RemoteApp" • enter https://remoteapp.math.uwaterloo.ca/RDWeb/Feed/webfeed.aspx • Start Menu -> "Work Resources (RADC)”

by: Jim Johnston

LINUX ENVIRONMENT

Brief overview of Linux machines:

• types of Linux machines

• aliases linux.math, biglinux.math and fastlinux.math

• GPU server, Rocks cluster, new CPU servers

• how to access them

• what they can be used for

by: Robyn Landers

WHY USE THE LINUX SERVERS

• Math applications/IDEs • Maple, Matlab, R, RStudio, Julia, CPLEX, Sage, NumPy, SciPy

• compilers and numerical libraries (gcc, GSL) • parallel tools on biglinux (OpenMP, MPI) • lots more memory than your laptop or Mac mini • long running jobs

JOB MANAGEMENT

• cpulimit

• screen

• checkpoint/restart with dmtcp

by: Christopher Calzonetti

What is fastlinux?:

• four computers with fastest CPUs, but only 2 CPUs per machine

• cpu127.math, cpu133.math, cpu135.math, and cpu137.math

• low core count

• not a large amount of memory

• What is fastlinux for?:

• jobs where processor speed is most important

Fastlinux

What is biglinux?:

• three computers, each with 4 high-core-count CPUs and large memory

• cpu125.math, cpu131.math and cpu139.math (cpu139 is newest, Xeon rather than Opteron)

What is biglinux for?:

• very large or multi-threaded jobs

Biglinux

ROCKS CLUSTER

• MPI programming across multiple machines • a cluster of eight two-CPU compute nodes • head node is mf-hpc.math.private • authenticate using Nexus password • home directory is research file server • submit jobs using SGE queue manager

JUPYTER• jupyter.math

GPU SERVERS

• gpu01.student.math • two 14-core CPUs, 128GB RAM • four NVIDIA Tesla K80 GPUs

• K80: ~5000 cores, ~2.9 teraFLOPS DP

• CUDA and other parallel GPU tools • authenticate using Nexus password • home directory is undergrad file

server (files.student.math)

• gpu101.math • two 14-core CPUs, 128 GB RAM • four NVIDIA Tesla P100 GPUs

• P100: ~3600 cores, 5.4 teraFLOPS DP

• CUDA and other parallel GPU tools • authenticate using Nexus password • home directory is research file

server (files.math)

ACCESS LINUX FROM YOUR DESKTOP

You will need: • uwuserid@uwaterloo.ca • password: same as your email password

From Mac mini - From Mac standalone - Go -> Utilities -> Terminal From Windows - putty or teraterm

Linux: open up a terminal window and type: ssh uwuserid@linux.math.uwaterloo.ca you will be prompted for your password

HPC (HIGH PERFORMANCE COMPUTING)

SHARCNET (www.sharcnet.ca) and Compute Canada (www.computecanada.ca)

• UW SHARCNET site has the new "Graham" cluster • 35,000 CPUs • some nodes with very large memory • 320 NVIDIA P100 GPUs • free accounts (need your supervisor to sponsor you)

SHARCNET offers • "how to" webinars • summer schools on HPC programming

https://uwaterloo.ca/mfcf/graduate-students

Happy Computing