C2115 Practical introduction to supercomputing Lesson 7 -1-
C2115
Practical introduction to
supercomputing
Lesson 6
Petr Kulhánek
kulhanek@chemi.muni.cz
National Centre for Biomolecular Research, Faculty of Science
Masaryk University, Kamenice 5, CZ-62500 Brno
C2115 Practical introduction to supercomputing Lesson 7 -2-
Content
➢ Computer architecture
CPU, memory, graphics system, disks, network, peripherals
➢ Submitting jobs
scripts vs programs, processes, nohup, screen, VNC
C2115 Practical introduction to supercomputing Lesson 7 -3-
Computer
architecture
C2115 Practical introduction to supercomputing Lesson 7 -4-
Overview
CPU Memory I/O
control bus
data bus
address bus
1945 von Neumannova architecture 1944 Hardwardskand architecture
CPU
Program
memory
Data
memory
I/O
John von Neumann, originally a
Hungarian mathematician working in
the United States
Harvard Mark I - computer made from
relays, 24 bit instructions
ro rw
rw
• the program can modify itself
• program and data cannot be loaded
at the same time
• the program cannot be modified
• the program and the data can be read at
the same time
C2115 Practical introduction to supercomputing Lesson 7 -5-
Overview
CPU Memory I/O
control bus
data bus
address bus
1945 von Neumannova architecture 1944 Hardwardskand architecture
CPU
Program
memory
Data
memory
I/O
John von Neumann, originally a
Hungarian mathematician working in
the United States
Harvard Mark I - computer made from
relays, 24 bit instructions
ro rw
rw
• the program can modify itself
• program and data cannot be loaded
at the same time
• the program cannot be modified
• the program and the data can be read at
the same time
in today's computers, both architectures are combined
C2115 Practical introduction to supercomputing Lesson 7 -6Typical
computer scheme
CPU
north
bridge
north
bridge
south
bridge
south
bridge
USB
mouse, keyboard
real time
clock
SATA controllers
hard drives
BIOS
graphics system
accelerator
memory
memory controller
peripherals with quick access via
PCI Express
Network (ethernet)sound
PCI bus
memory controller becomes part
of the latest processors
C2115 Practical introduction to supercomputing Lesson 7 -7-
CPU
Processor also CPU (Central Processing Unit) is an essential part of a computer; It is a very
complex sequential circuit that executes machine code stored in the computer's operating
memory. The machine code consists of individual machine instructions of computer
programs loaded into the operating memory. www.wikipedia.org
CPU
controller
ALU
ALU (arithmetic and logic unit),
performs arithmetic operations,
evaluates conditions
reads machine instructions and data
and prepares their processing in ALU
sequential processing of machine instructions is
controlled by an internal clock cycle
C2115 Practical introduction to supercomputing Lesson 7 -8Computational
tasks
C2115 Practical introduction to supercomputing Lesson 7 -9Jobs
vs Processes
HW
CPU, RAM, HDD,…
OSOS
User space
process process
process
p
p
Multitasking, in informatics, indicates ability of the operating
system to perform several processes simultaneously (at least
seemingly). The core of the operating system very quickly
changes processes running on the processor or processors (socalled
context change), so that the computer user has the
impression that they are all running at the same time.
Process, in informatics, is the name for a running computer
program. The process is located in the computer operating
memory as sequence of machine instructions executed by
processor. Process management is performed by operating
system, which ensures their separate running, allocates them
computer system resources, and allows the user to manage
processes (start, end, etc.).
edited from wikipedia.org
A computational job is a process or group of processes.
C2115 Practical introduction to supercomputing Lesson 7 -10Programs
vs Scripts
source code
program
input
output
translation
(compilation)
script interpreter
input
output
Program is a set of machine instructions
processed directly by the processor. The
program is created by translating source
code of the programming language.
Translated languages:
• C / C ++
• Fortran
Script is a text file containing commands
and control sequences that are executed
by interpreter of used scripting language.
Scripting languages:
• bash
• gnuplot
• awk
• JavaScript
• PHP
(Process) (Process)
C2115 Practical introduction to supercomputing Lesson 7 -11Programs
vs Scripts
Programs designed for demanding scientific and technical calculations are always written
in compilable programming languages. These languages ​​include:
➢ Fortran
➢ C/C ++
Scripting languages are either not used at all for demanding calculations or are used in
supporting parts of the calculations which are not computationally intensive.
C2115 Practical introduction to supercomputing Lesson 7 -12-
Commands
top continuously displays processes sorted by CPU load (end with q key)
ps lists processes running in given terminal or according to provided
specification
(ps -u user_name)
pstree lists processes (tree listing)
type lists path to the standard application/command
Kill sends a signal to process, can be used to terminate problematic programs
time lists length of the process run
ssh runs command on remote computer
sleep starts a process that waits for specified time
& runs process in background
wait waits for background processes to complete
nohup starts the process without interacting with the terminal
screen screen terminal multiplexer
vncserver starts VNC server
vncwiever connects to VNC server } tigervnc
C2115 Practical introduction to supercomputing Lesson 7 -13Exercise
1
1. Determine the percentage of programs written in Fortran, C/C++, and others that are
listed on the following page:
http://en.wikipedia.org/wiki/List_of_quantum_chemistry_and_solid_state_physics_software
Display the result in the form of a pie chart.
Note: use the grep and wc commands to solve the exercise.
C2115 Practical introduction to supercomputing Lesson 7 -14Submitting
jobs
➢ nohup
➢ screen
➢ tigervnc
C2115 Practical introduction to supercomputing Lesson 7 -15Exercise
2
1. Log in to the wolf39 node in two terminals. In one of them, run the command:
2. Run pi program (it is part of the system) in the second terminal. What is it used for?
3. Write a script that calculates the number pi with an accuracy of 10 to 10,000 digits.
Save the resulting pi numbers into the pi.txt file.
4. Run the script in the background and log out of the wol39 node.
5. What will happen?
6. Log in to the wolf39 node again. Kill the running job.
7. Modify the script so that the result is printed to standard output. Run the script again,
this time in the foreground.
8. What happens if the network connection is lost? What is the purpose of the SIGHUP
signal?
$ top -u <login_name>
C2115 Practical introduction to supercomputing Lesson 7 -16Running
jobs
To run a job that is immune to termination by the SIGHUP signal, you can use:
• nohup (for non-interactive tasks)
• screen or tmux (for interactive tasks)
• VNC session (for GUI tasks)
• batch system
Commands nohup, screen, tmux are used only to run jobs on computers where a batch
system is not available.
In supercomputer centers, these commands can only be used in batch jobs (screen, VNC)
or to run service tasks (which are not computationally intensive) on the front nodes.
C2115 Practical introduction to supercomputing Lesson 7 -17-
nohup
Typical use:
$ nohup ./my_script &
Script output my_script attaches to the file nohup.out.
Note:
• Processes running in the background (started with nohup) may be terminated by the
system administrator after the last user session is closed.
• This can be prevented by changing the configuration:
/etc/systemd/login.conf
[Login]
KillUserProcesses=no
C2115 Practical introduction to supercomputing Lesson 7 -18-
screen
Typical use:
$ screen
win1$ ./my_script
win1$ # Ctrl+a d – leaves session
$ screen –list # lists open sessions
$ screen –r session.id # connects into session
Note:
• Background processes (screen and processes running in its instance) can be terminated
by the system administrator after the last user session is closed.
• This can be prevented by changing the configuration:
/etc/systemd/login.conf
[Login]
KillUserProcesses=no
C2115 Practical introduction to supercomputing Lesson 7 -19Exercise
3
1. Log in to node wolf39.
2. Open screen program session. Run the script to calculate the number pi in it.
3. Disconnect from the session and log out of the wolf39 node.
4. Log in to node wolf39 and resume the session in the program screen. What are you
observing?
C2115 Practical introduction to supercomputing Lesson 7 -20-
VNC
Virtual Network Computing (VNC) is a graphical program that allows remote connection to
a graphical user interface using a computer network.
VNC works as a client-server, where the server creates a graphical desktop in the computer
operating memory and communicates over the network with the client, which displays the
desktop to the user (usually on another computer).
Remote Frame Buffer protocol (RFB) is used for communication and the aim is to minimize
the volume of data transferred between the client and the server and thus enable
communication even over slower data lines (e.g., via the Internet).
wikipedia.cz
VNC does not offer secure data transfer by default!!!!
Note:
• Processes running (vncserver) in the background can be terminated by the system
administrator after the last user session.
• This can be prevented by changing the configuration:
/etc/systemd/login.conf
[Login]
KillUserProcesses=no
C2115 Practical introduction to supercomputing Lesson 7 -21VNC
- transmission security
PC
ssh tunnel
DESKTOP
DESKTOP
vncserver
vncviewer
unix socket, access is
restricted to users
remote linux
machine
TCP port > = 5900
firewall
Securing the transmission is quite complicated. Either requiring the creation of an X.509
certificate or creating a a secure line using ssh tunneling.
The second option is offered by modified versions of vncviever and vncserver programs
from the module tigervnc.
unix socket, access is
restricted to users
accessible to local users
C2115 Practical introduction to supercomputing Lesson 7 -22-
tigervnc
wolf39$ module add tigervnc
wolf39$ vncserver
..
VNCID: kulhanek@wolf39.ncbr.muni.cz:1
..
wolf39$
wolf01$ module add tigervnc
wolf01$ vncviewer kulhanek@wolf39.ncbr.muni.cz:1
server
client
1
2
C2115 Practical introduction to supercomputing Lesson 7 -23Exercise
4
1. On node wolf39, run vncserver from the module tigervnc and connect to it from your
workstation. In the VNC session, run the VMD program and open the chitin nanofiber
model. Leave the seating open.
2. Log in to the wolf39 node and run the VMD program on it (using the display export)
and display the chitin nanofiber model again.
3. Compare the speed of interactive work with the VMD program, which displays using
VNC and directly. Discuss the observations.