C2115 Practical Introduction to Supercomputing 6th Lesson -1C2115
Practical Introduction to
Supercomputing
Petr Kulhánek, Jakub Štěpán
kulhanek@chemi.muni.cz
National Centre for Biomolecular Research, Faculty of Science
Masaryk University, Kotlářská 2, CZ-61137 Brno
CZ.1.07/2.2.00/15.0233
6th Lesson
C2115 Practical Introduction to Supercomputing 6th Lesson -2-
Contents
 Multiuser environment
process, thread, multitasking, context switch
 Exercise
parallel application run efficiency
C2115 Practical Introduction to Supercomputing 6th Lesson -3Multiuser
environment
 Process
 Thread
 Multitasking
 Context switch
C2115 Practical Introduction to Supercomputing 6th Lesson -4Process
and thread
Process is in informatics name of running computer program. Process is placed in
operating memory as a machine instruction sequence processed by processor. Contains
not only code of processed program, but also dymanically changing data, that are
processed. One program may run as multiple processes with different data. Process
management is done by operating system, that ensures their separate run, assigns system
resources and provides user tools to manage processes.
Thread is lightweighted process, that allows lowering down costs of operating system
during context switch, that is necessary to allow massive parallel calculations. While
processes are strictly separated, threads share same memory space and other structures.
One process may handle multiple threads, threads comunicate easily through shared
memory, but this brings also possible problems in concurent memory access - race
condition.
source: www.wikipedia.cz, adjusted
C2115 Practical Introduction to Supercomputing 6th Lesson -5-
Multitasking
Multitasking is ability of operating system to seemingly process multiple processes at one
time. Operating system core switches running processes on processor very fast (context
switch), so these processes seem to be running concurently.
Preemptive multitasking means that operating system itself decides about resources
assigned to particular processes. Periodically (typical rate is approximately 100× to 1000×
per second) interupts processing of running process, evaluates situation (number of waiting
processes, their priorities etc.) and then decides either to start interupted process again or
starts onother process that was waiting. During process switch also context has to be
switched. Process in preemptive multitasking may ask for context switch and give up its
resources (it is put to „sleep“ or waits for slow input-output operatins, as is for example
hard disk reading).
source: www.wikipedia.cz, adjusted
Types:
non-preemptive multitasking (not used much nowadays)
preemptive multitasking (modern OS)
C2115 Practical Introduction to Supercomputing 6th Lesson -6Context
switch
Context
The term denotes processor state (register contents), state of coprocessor and possibly
other devices in time of context change. This particular state is saved either to process
stack or to dedicated part of process memory space.
Contaxt contains also contents of processor cache memory levels (for example L1 cache or
TLB): these are not saved, but their contents are implicitly or explicitly invalidated during
context switch. Necessity of their new loading is main reason why context switch is so
time demanding on modern architectures.
Context switch is operation of multitaking operating system that switches controll among
processes. This implies saving and loading of current processor state. This is repeated
many times per second. Context changes are usually computationaly intensive.
source: www.wikipedia.cz, adjusted
C2115 Practical Introduction to Supercomputing 6th Lesson -7-
Exercise
 Parallel application run efficiency
C2115 Practical Introduction to Supercomputing 6th Lesson -8Exercise
LV.1
1. How many processes contains your machine? Get machine mane, CPU type and
number. Use command lscpu.
2. Extend and compare data obtained in task 1 with information from file /proc/cpuinfo.
3. What is size of cache memory L1, L2 and L3 CPU in your machine?
4. Get your CPU frequency?
5. Compile program load_cpu.f90, that is in directory
/home/kulhanek/Data/C2115/programs
6. State runtime of program load_cpu using command:
7. Give reason why not to use directly command time? What is option --format?
$ gfortran -O3 load_cpu.f90 -o load_cpu -lblas
$ /usr/bin/time --format=%e ./load_cpu
C2115 Practical Introduction to Supercomputing 6th Lesson -9Exercise
LV.2
1. To run program load_cpu use following script in bash.
2. Do script function analysis.
3. Measure script runtime. Measured time compare with theoretical expectations based
on run of sigle program load_cpu. Do measurements for
N=1,NCPU,2*NCPU,3*NCPU,4*NCPU,5*NCPU,6*NCPU, where NCPU is CPU number
accesible on your machine. Monitor running processes by command top in separate
terminal.
4. Analyze obtained differences in runtime and explain them.
#!/bin/bash
N=4 # parralel runs number
for((I=1;I<=N;I++)); do
./load_cpu & # runs application on backgroud
done
wait # waits for all background runs to finish