MOSIX tutorial

Amnon Barak

Description

MOSIX TutorialL. Amar, A. Barak, T. Maoz, E. Meiri, A. Shiloh Department of Computer ScienceThe Hebrew University http:// www . MOSIX . orgCopyright © Amnon Barak 20101Copyright © Amnon Barak 2010AboutThis tutorial has 2 parts:• Part I: is for new users. It covers basic tools and operations such as monitors, how to run and control processes, view processes, view and control the queue and how to handle unsupported features• Part II: includes advanced topics such as freezing processes, checkpoint & recovery, running a large set of processes, I/O optimizations, running Matlab, running parallel jobs, configuration and management, and the programming interface2Copyright © Amnon Barak 2010Part I: Basics• Tools• Operations– mon, mmon– Initial assignment– mosrun, native– Running Linux – mospsprocesses– mosq– migrate• Encountering unsupported features Detailed information about each command is available in the corresponding manual pages:man mosix | mosrun | mosps…3Copyright © Amnon Barak 2010Monitors - seeing what is going onmon and mmon are 2 monitors for viewing the status of the resources in each cluster and all the clusters mon – displays basic information (tty format) about resources in the local cluster To display type:  “l” – CPU load (relative)  “f” – Number of frozen processes  “m” - Memory (used + free), swap-space (used + free) – type consecutively  “u” - Utilization  “d/D” - Dead nodes  “h” – ...

Informations

• Publié par: Thiwyer
• Nombre de lectures: 364
• Langue: English

Extrait

MOSIX Tutorial

L. Amar, A. Barak, T. Maoz, E. Meiri, A. Shiloh Department of Computer Science The Hebrew University

http:// www . MOSIX . org

Copyright © Amnon Barak 2010

Copyri

About

This tutorial has 2 parts: •Part I: is for new users. It covers basic tools and operations such as monitors, how to run and control processes, view processes, view and control the queue and how to handle unsupported features •Part II: includes advanced topics such as freezing processes, checkpoint & recovery, running a large set of processes, I/O optimizations, running Matlab, running parallel jobs, configuration and management, and the programming interface

ght ©Amnon Barak  2010

2

Copyri

Part I: Basics

•Tools   – mon, mmon – mosrun, native –mosps –mosq –migrate

ght ©Amnon Barak  2010

•snopiOtrae –Initial assignment –Running Linux processes

•erinountEnc g unsupported features

Detailed information about each command is available in the corresponding manual pages: man mosix | mosrun | mosps…

3

aBaron n100 k2  eingsesi tahw o gniog n CyrophtigAm ©

monandmmonare 2 monitors for viewing the status of the resources in each cluster and all the clusters mon–displays basic information (tty format) about resources in the local cluster   To display type:   “l”– CPU load (relative)    “f”– Number of frozen processes    “m”-    (used + free) –Memory (used + free), swap-spacet   ype cutivel   conse y “u”- Utilization    “d/D”- Dead nodes    “h”– help for complete list of options    

4

Monitors -

tivity (twork aco“ ” –en laI O/       e  ” “s PC Rat rtot ,)la ,ni,tuo versSIX , MOatus  tsee,d Ups -PC s  denod ea d -D”/d“    emitpu ,noi ,ed si-kpsca  e “   ree),  swap-spac s reutatr( sinnu –g”ul m-ctistlurotip ir i“ ” )y    ocalng lsts,/gue r etirW tot ,eta IadRe– e,at r/O h”   “ h –p elr fopmocetelsil fo t options

mmon– includes all the options ofmonand many more   To display type (consecutively):  

5

mmon

010raBa 2k Am ©n nogithporyC Memory (used + f orecssse   m“ ” - ssce, esozfr pen daorf +nezoorp ” -    “l,  lLoad

Copyrig

Othermmon

features

Can run on non-MOSIX nodes, e.g. your workstation   mmon –h bmos- 01(node #1 in the bmos cluster)    Display the status of several clusters (consecutively)   Example:mmon –c amos- 01 , bmos- 01 , cmos- 01     ter to anoth keys to skip from one clUse the > ,< us er Pr   ivate color scheme using the file~/.mmon.cfg  

ht ©Amnon Barak  2010

6

mosrun– r

To run a program under the MOSIX discipline start it   withmosrun,e.g.,mosrun myprog Such programs can migrate to other nodes    Example:    1 2 3> mosrun myprog(run myprog, possibly with arguments) Programs thatare not started bymosrunrun in native   Linux mode andCANNOTmigrate A program that is started bymosrunand all its children   remain under the MOSIX discipline MOSIX processes (that were started bymosrun)can use thenativeutility to spawn children that run in native   Linux mode

7

opyrC ©AmightaBaron n100 k2  nnug inISOMrp Xsecoses

Copyrig

Example: view the process migration

        

  

  

ht ©A

Login to any node in a MOSIX cluster

On one window runmonormmon

On another window start 2 CPU-intensive processes, e.g. the testloadprogram (included in the MOSIX distribution):

>mosrun testload &

>mosrun testload &

Observe in themon/mmonwindow how the processes move across the cluster’s nodes

Typemoskillallto stop the running processes

mnon Barak  2010

8

mosrun– n

mosrun –rmos 1 myprog(run on node mos 1 ) > > mosrun –b myprog(run on the best node) > mosrun –3 myprog 1 2 3(run on node #3, with arg 1 2 3)

  

Examples:

-jID1-ID2[,ID3-ID4]on a random node in the ranges– run

ID1-ID2, ID3-ID4, …

-h– run on thehomenode

-b– attempt to select the best node

-{a.b.c.d}– run on node with thisIP

-{n}– run on node numbern

                 

-r{hostname}– run on this host (node)

9

sa engistnemtpo nsio  tA©nmnoB raka  2010 od Copyrigh

mosrun– w

-G– allows processes to migrate to nodes in other clusters   Otherwise, processes are confined to the local cluster    -G{class}– if class > 0 than a process is allowed to migrate to nodes in other clusters. Note that–Gi   s equivalent to–G1   tysegebam-m{} – specifies the maximal amount of memory   needed by your program, to prevent migration of processes to nodes that do not have sufficient free memory Beside migration, the–Gand-moptions also affect the initial assignment (-bflag) and queuing (see below)   Example:   > mosrun –G –m1000 myprog(allowsmyprogto run on other clusters, but only on nodes with at least 1GB of free memory)

10

CyropBaranon  ©Amight100 k2  esocprn cae erhot etargim ses

 k2 aBar100ightopyrnon  ©AmC

mosrun-

11

The-J{Job Id}option ofmosrunallows bundling for easy identification of several instances ofmosrun   The“Job Id”is an integer (default value is 0)    Each user can assign their own“Job Ids”    “Job Id”i   s inherited by all child processes    “Job Id”c   an be viewed bymosqandmosps    All jobs of a user with the same“Job Id”c   an be collectively   killed (signaled) bymoskillalland migrated bymigrate Examples:   > mosrun –J20 myprog(runmyprogwith Job_ID = 20) > mosps –J20(   list only my processes with Job_ID = 20) > moskillall –J20(kill all my processes with Job_ID = 20)

oyruni gamkrescess pro