Job Submission Slides for Grid Computing Techniques and

Job Submission Slides for Grid Computing: Techniques and Applications by Barry Wilkinson, Chapman & Hall/CRC press, © 2009. Chapter 2, pp. 35 -59. For educational use only. All rights reserved. Aug 24, 2009 2 -1. 1

Types of jobs to be submitted to a Grid • Programs written in C, C++, … that need to be compiled. • Java programs that need a Virtual Java Machine • Pre-compiled application packages 2 -1. 2

Submitting a job that needs to be compiled Fig. 2. 1 2 -1. 3

Java programs • Quite similar to compiling C programs, except Java compiler (javac) creates class file (bytecode) that is interpreted by a Java Virtual Machine (java). • It is the Java Virtual Machine that is the executing program and the class file is an input file. • Other class files usually need to be called too, found in path specified by CLASSPATH variable, so this variable must be set up properly. 2 -1. 4

Submitting a Java job Fig. 2. 2 2 -1. 5

• Java programs offer more portability because class file could be sent to any remote computer having a Java Virtual Machine installed. • However, speed of execution may be less than executing fully compiled binaries. • Some studies have shown Java programs to run at 70% of equivalent C programs. • Many internal components of Grid middleware software such as Globus actually use a mixture of Java and C. Java commonly used to create Web service components. 2 -1. 6

Types of Applications Since Grid is a collection of computers, user might wish to use these computers collectively to solve problems. Two ways: • Parallel programs -- Break problem down into tasks that need to be done to solve problem and submit individual tasks to different computers to work on them simultaneously. • Parameter sweep problems -- Run same job on different computers at same time but with different input parameters. Particularly attractive for Grid computing platforms because no dependences between each sweep (usually). 2 -1. 7

Grid Resource Allocation Management (GRAM) Principal job submission component of Globus 2 -1. 8

Globus Open Source Grid Software G T 4 G T 3 G T 2 G T 3 G T 4 Community Scheduler Framework [contribution] Delegation Service Python WS Core [contribution] C WS Core GRAM Community Authorization Service OGSA-DAI [Tech Preview] WS Authentication Authorization Reliable File Transfer Grid Resource Allocation Mgmt (WS GRAM) Monitoring & Discovery System (MDS 4) Java WS Core Grid. FTP Grid Resource Allocation Mgmt (Pre-WS GRAM) Monitoring & Discovery System (MDS 2) C Common Libraries Pre-WS Authentication Authorization Web Services Components Non-WS Components Replica Location Service XIO Credential Management Security Data Management Execution Management Information Services Common Runtime I Foster

Job submission components Fig. 2. 3 2 -1. 10

Running simple jobs across a Grid computing environment Fig. 2. 4 2 -1. 11

Specifying the job Two basic ways a job might be specified: • Directly by name of executable with required input arguments or • By a job description file – more powerful 2 -1. 12

Directly For very simple jobs, one can submit a single job using -c option, e. g. , globusrun-ws -submit -c prog 1 arg 2 which executes program prog 1 with arguments arg 1 and arg 2 on local host. -c option actually causes globusrun-ws to generate a job description with the named program and arguments that follow. -c option must be the last globusrun-ws option (why? ). 2 -1. 13

Example globusrun-ws –submit –c /bin/echo hello Globus job monitoring output created on command line and will indicate that the job completes. However, output from echo program (hello) not displayed and is lost as is any standard output without further specification (see later). 1 b. 14

Job Description File Gives details such as: • Job Description - Name of executable - Number of instances - Arguments - Input files - Output files - Directories - Environment variables, paths, . . . • Resource requirements - Processor - Number, cores, . . . - Type - Speed, . . . - Memory Used to match job with resources 2 -1. 15

Job Description Languages Several languages invented. • Globus - specific: – Globus 1 and 2 used their Resource Specification language RSL (version 1) – Globus 3 used an XML version called RSL-2 – Globus 4 uses a variation of RSL-2 in a JDD (Job Description Document) • Job Submission Description Language (JSDL) – A recent industry-wide standard (2005) 2 -1. 16

Resource Specification Language RSL version 1 • A meta-language describing job and its required execution. Provides specification for: • Job description - directory, executable, arguments, environment • Resource requirements - machine type, number of nodes, memory, etc. 2 -1. 17

RSL Version 1 examples Constraints Example Conjunction (AND): & • To create 3 -5 instances of my. Prog, each on a machine with at least 64 Mbytes memory available to me for 1 hours: & (executable=my. Prog) (count>=3)(count<=5)(memory>=64) (max_time=60) 2 -1. 18

Constraints Example Disjunction (OR): | • To create 5 instances of my. Prog, each on a machine with at least 64 Mbytes memory or 7 instances of my. Prog, each on a machine with at least 32 Mbytes memory : &(executable=my. Prog) (|(&(count=5)(memory>=64)) (&(count=7)(memory>=32))) 2 -1. 19

Requesting multiple resources multirequest: + • To execute 5 instances of my. Prog 1 on a machine with at least 64 Mbytes memory and execute 2 instances of my. Prog 2: +(&(count=5)(memory>=64)) (executable=my. Prog 1)) (&(count=2)(executable=my. Prog 2)) 2 -1. 20

XML Job Description languages • With introduction of XML in early 2000’s, job description languages began to be changed to XML. 2 -1. 21

Using XML • Much more elegant and flexible, and in keeping with Web services. • Can use XML parsers. • Allows more powerful mechanisms with job schedulers. • Resource scheduler/broker applies specification to local resources. 2 -1. 22

Resource Specification Language, RSL version 2 • XM job description language used Globus version 3 (GT 3). • An XML language. 2 -1. 23

RSL-2 • XML version of RSL 1 • Can specify everything from executable, paths, arguments, input/output, error file, number of processes, max/min execution time, max/min memory, job type etc. 2 -1. 24

GT 3 RSL-2 Example Specifying Executable (executable=/bin/echo) <gram: executable> <rsl: path> <rsl: string. Element value="/bin/echo"/> </rsl: path> </gram: executable> 2 -1. 25

• • • • • • • • • • • • • RSL and GT 3. 2 RSL-2 comparison for echo program <? xml version="1. 0" encoding="UTF-8"? > <rsl: rsl xmlns: rsl="http: //www. globus. org/namespaces/2003/04/rsl" xmlns: gram="http: //www. globus. org/namespaces/2003/04/rsl/gram" xmlns: xsi="http: //www. w 3. org/2001/XMLSchema-instance" xsi: schema. Location=" http: //www. globus. org/namespaces/2003/04/rsl c: /ogsa-3. 0/schema/base/gram/rsl. xsd http: //www. globus. org/namespaces/2003/04/rsl/gram c: /ogsa-3. 0/schema/base/gram_rsl. xsd"> <gram: job> <gram: executable> <rsl: path> <rsl: string. Element value="/bin/echo"/> </rsl: path> </gram: executable> <gram: directory> <rsl: path> <rsl: string. Element value="/bin"/> </rsl: path> </gram: directory> <gram: arguments> <rsl: string. Element value="Hello World"/> </rsl: string> </gram: arguments> <gram: stdin> <rsl: path> <rsl: string. Element value="/dev/null"/> </rsl: path> </gram: stdin> <gram: stdout> <rsl: path. Array> <rsl: path> <rsl: substitution. Ref name="HOME"/> <rsl: string. Element value="/stdout"/> </rsl: path. Array> </gram: stdout> <gram: stderr> <rsl: path. Array> <rsl: path> <rsl: substitution. Ref name="HOME"/> <rsl: string. Element value="/stderr"/> </rsl: path. Array> </gram: stderr> <gram: count> <rsl: integer value="1"/> </gram: count> <gram: job. Type> <gram: enumeration. Value> <gram: multiple/> </gram: enumeration. Value> </gram: enumeration> </gram: job. Type> <gram: gram. My. Job. Type> <gram: enumeration. Value> <gram: collective/> </gram: enumeration. Value> </gram: enumeration> </gram: gram. My. Job. Type> <gram: dry. Run> <rsl: boolean value="false"/> </gram: dry. Run> <gram: save. State> <rsl: boolean value="true"/> </gram: save. State> <gram: two. Phase> <rsl: integer value="600"/> </gram: two. Phase> </gram: job> </rsl: rsl> &((executable=/bin/echo) (directory="/bin") (arguments="Hello World") (stdin=/dev/null) (stdout="stdout") (stderr="stderr") (count=1) ) 2 -1. 26

Job Description Document (JDD) • RSL-2 renamed and called JDD used in more recent Globus 4 (GT 4) documents. • Similar to original RSL-2 but simplified syntax. • Not completely interchangeable. 2 -1. 27

GT 4 JDD Example Specifying Executable executable=/bin/echo <executable>/bin/echo</executable> 2 -1. 28

GT 4. 0 JDD for echo program <? xml version="1. 0" encoding="UTF-8"? > <job> <executable>/bin/echo</executable> <directory>${GLOBUS_USER_HOME}</directory> <argument>Hello</argument> <argument>World</argument> <stdout>${GLOBUS_USER_HOME}/stdout</stdout> <stderr>${GLOBUS_USER_HOME}/stderr</stderr> </job> 2 -1. 29

Job Submission Description Language (JSDL) • A standard introduced by GGF (Global Grid forum) in 2005 and beginning to be widely adopted. 2 -1. 30

Basic JSDL structure <Job. Definition> <Job. Description> <Job. Identification >. . </Job. Identification> <Application>. . . </Application> <Resources>. . . </Resources > <Data. Staging>. . . <Data. Staging > </Job. Description> </Job. Definition> 2 -1. 31

For executables operating in a Linux environment, replace <application> with <POSIXapplication> <POSIXApplication name=”xsd: . . . ”> <Executable>. . . </Executable> <Argument>. . . </Argument> <Input>. . . </Input> <Output>. . . </Output> <Error>. . . </Error> <Working. Directory>. . . </Working. Directory> </POSIXApplication> Portable Operating System Interface, a collection of IEEE standards that define APIs, compatible to most versions of Unix/Linux 2 -1. 32

Sample Linux job description <? xml version="1. 0" encoding="UTF-8"? > <jsdl: Job. Definition xmlns: jsdl="http: //schemas. ggf. org/jsdl/2005/11/jsdl" xmlns: jsdl-posix="http: //schemas. ggf. org/jsdl/2005/11/jsdl-posix"> <jsdl: Job. Description> <jsdl: Application> <Job. Name>Test Job</Job. Name> <Description>Hello world Job</Description> <jsdl-posix: POSIXApplication > <jsdl-posix: Executable>/bin/echo</jsdl-posix: Executable> <jsdl-posix: Argument>hello, world</jsdl-posix: Argument> <jsdl-posix: Output>${GLOBUS_USER_HOME}/stdout</jsdlposix: Output> <jsdl-posix: Error>${GLOBUS_USER_HOME}/stderr</jsdlposix: Error> </jsdl-posix: POSIXApplication> </jsdl: Job. Description> </jsdl: Job. Definition> 2 -1. 33

<Resources> describes requirements of resources for job and can include: <Resources> <Candidate. Hosts>. . . </Candidate. Hosts> <File. System>. . . </File. System> <Exlusive. Execution>. . . </Exlusive. Execution> <Operating. System>. . . </Operating. System> <CPUArchitecture>. . . </CPUArchitecture> <Individual. CPUSpeed>. . . </Individual. CPUSpeed> <Individual. CPUTime>. . . </Individual. CPUTime> <Individual. CPUCount>. . . </Individual. CPUCount> <Individual. Network. Bandwidth>. . . </Individual. Network. Bandwidth> <Individual. Physical. Memory>. . . </Individual. Physical. Memory> <Individual. Virtual. Memory>. . . </Individual. Virtual. Memory> <Individual. Disk. Space>. . . </Individual. Disk. Space> <Total. CPUTime>. . . </Total. CPUTime> <Total. CPUCount>. . . </Total. CPUCount> <Total. Physical. Memory>. . . </Total. Physical. Memory> <Total. Virtual. Memory>. . . </Total. Virtual. Memory> <Total. Disk. Space>. . . </Total. Disk. Space> <Total. Resource. Count>. . . </Total. Resource. Count> </Resources> 2 -1. 34

Submitting a job 2 -1. 35

GT 4 job submission command globusrun-ws • Submit and monitor GRAM jobs • Written in C, for faster startup and execution than earlier Java version • Supports multiple and single job submission • Handles credential management • Streaming of job stdout/err during execution 2 -1. 36

Simple job submission • Step 1: Create proxy with: grid-proxy-int command. • Step 2: Issue globusrun-ws with parameters to specify job. 2 -1. 37

Some globusrun-ws flags (options) for job submission 2 -1. 38

Running GT 4 Job using XML job description file • Command: globusrun-ws –submit –f prog. xml where prog. xml specifies job in JDD. -submit causes job to be submitted Submitted to localhost (machine that is executing command) as no contact resource specified. Submitted immediately using “fork” 2 -1. 39

With named executable -c option Example: Submit program echo with argument hello to default localhost. globusrun-ws –submit –c /bin/echo hello -c Causes globusrun-ws to generate job description with named program and arguments. -c option, if used, must be last option. Only useful for very simple single jobs. 2 -1. 40

Output modes -submit Submits (or resubmits) a job in one of three output modes: batch interactive, or interactive-streaming. Default (without additional flags to specify) is interactive. 2 -1. 41

Interactive mode Example Submit program echo with argument hello to default localhost. % globusrun-ws –submit –c /bin/echo hello Submitting job. . . Done. Job ID: uuid: d 23 a 7 be 0 -f 87 c-11 d 9 -a 53 b 0011115 aae 1 f Termination time: 07/20/2005 17: 44 GMT Current job state: Active Current job state: Clean. Up Job goes thro several states Current job state: Done Destroying job. . . Done. Job ID Output 2 -1. 42

Streaming Refers to sending contents of a stream of data from one location to another location as it is generated. Often associated with Linux standard output and standard error streams, stdout and stderr. For a program that creates output on remote machine, need: • Files to hold output and error messages , or • Re-direct output and error messages to user console. 2 -1. 43

Interactive-streaming mode -s option Provides for capturing program output and error messages and re-directing them to user’s console (output of globusrun-ws) or to specified files. 2 -1. 44

Interactive-streaming mode Re-direction to user console -s option Example globusrun-ws -submit -s -c /bin/echo hello Output (hello) redirected to (globusrun-ws) stdout Error messages redirected to (globusrun-ws) stderr 2 -1. 45

Interactive-streaming mode Re-direction to files -s option with –so and –se options -s for streaming output and –so to specify output file –se to specify error file 2 -1. 46

Example globusrun-ws -submit -s -so outfile -se errorfile -c /bin/echo hello name of file holding output Argument for echo name of file holding error messages 2 -1. 47

Specify streaming to files using Job description file Example (JDD) <job> <executable>/bin/echo</executable> <argument>Hello</argument> <stdout>job. Out</stdout> <stderr>job. Err</stderr> </job> 2 -1. 48

Batch submission A long-standing Computer Science term from early days of computing where jobs submitted to system in a group (a batch) and wait their turn to be executed sometime in the future. Originally appeared when programs were submitted by punched cards to a shared system, perhaps to be run perhaps overnight. (The author remembers those days with frustration. ) Batch submission really part of a scheduling approach. 2 -1. 49

Batch submission -b option In globusrun-ws, batch referred to as an output mode because of way output generated. Once job submitted, control returned to command line, and one will need to query system to find out status of job. 2 -1. 50

For example, suppose we ran the job: globusrun-ws –submit /bin/sleep 100 in interactive mode. Would return when program (sleep for 100 seconds in this case) completes. We would get normal globusrun-ws output, such as: Submitting job. . . Done. Job ID: uuid: d 23 a 7 be 0 -f 87 c-11 d 9 -a 53 b-0011115 aae 1 f Termination time: 07/20/2005 17: 44 GMT Current job state: Active Current job state: Clean. Up Current job state: Done Destroying job. . . Done. only each line would appear as process moves to next status condition. 2 -1. 51

Alternatively, could execute sleep in batch output mode: (-b option): globusrun-ws –submit –b /bin/sleep 100 Output would immediately appear of the form: Submitting job…Done Jo. B ID: uuid: f 9544174 -60 c 5 -11 d 9 -97 e 3 -0002 a 5 ad 41 e 5 Termination time: 01/08/2005 16: 05 GMT Displays Managed. Job EPR as job ID (more on this later). Control returned to command line. Program may not have finished. In this case it will not for 100 seconds. 2 -1. 52

Now one has to query state of job to find out when it completes. Need job ID (Managed. Job EPR) Convenient to have that put in a file using –o option when submitting job, e. g. globusrun-ws –submit –b -o job. EPR /bin/sleep 100 where job. EPR holds the job ID (Managed. Job EPR). 2 -1. 53

To watch status of submitted job “Attach” interactive monitoring with -monitor option. Job ID (Managed. Job EPR) provided with -j option, e. g. : globusrun-ws –monitor –j job. EPR where job. EPA holds Managed. Job EPR. Then can see stages job goes through with interactive output immediately: job state: Active Current job state: Clean. Up Current job state: Done Requesting original job description. . . Done. Destroying job. . . Done although job itself still batch output job. 2 -1. 54

Some other options -status Reports the current state of the job and exits -kill Requests immediate cancellation of job and exits. 2 -1. 55

2 -1. 56

Specifying where job is submitted Request to run job processed by “factory” service called Managed. Job. Factory. Service. Default URL: https: //localhost: 8443/wsrf/services/Managed. Job. Factory. Service 2 -1. 57

To specify where job is submitted -F Specifies “contact” for the job submission. globusrun-ws –submit –F http: //localhost: 8440 –f prog 1. xml Job submitted to localhost Globus container that hosts services running on port 8440 Factory service still located at. wsrf/services/Managed. Job. Factory. Service 2 -1. 58

Selecting a different host Example globusrun-ws –submit –F https: //140. 221. 65. 193: 4444/wsrf/ services/managed. Job. Factory. Service –f prog 1. xml 2 -1. 59

Many other options Example -term time Set an absolute termination time, or a time relative to successful job creation 2 -1. 60

Transferring Files Job submission command, for example: globusrun-ws –submit –F http: //coit-grid 01. uncc. edu –c prog 1 requires prog 1 to be existing on the remote machine in the default directory ( ${GLOBUS_USER_HOME} ). Up to user to ensure executable is in place. 2 -1. 61

Grid. FTP A Globus component that provides for: • Large data transfers • Secure transfers • Fast transfers – Parallel transfers -- employing multiple virtual channels sharing a single physical network connection – Striping -- employing multiple physical channels using multiple hardware interfaces. • Reliable transfers • Third party transfers. 2 -1. 62

Globus Open Source Grid Software G T 4 G T 3 G T 2 G T 3 G T 4 Community Scheduler Framework [contribution] Delegation Service Python WS Core [contribution] C WS Core Community Authorization Service OGSA-DAI [Tech Preview] WS Authentication Authorization Reliable File Transfer Grid Resource Allocation Mgmt (WS GRAM) Monitoring & Discovery System (MDS 4) Java WS Core Grid. FTP Grid Resource Allocation Mgmt (Pre-WS GRAM) Monitoring & Discovery System (MDS 2) C Common Libraries Pre-WS Authentication Authorization Non-WS Components Replica Location Service XIO Grid. FTP Credential Management Security Web Services Components Data Management Execution Management Information Services Common Runtime I Foster

Third party transfers Transferring a file from one remote location to another remote location controlled by a party at another location (the third party). Already seen third party transfers in Grid portal at file management portlet. There, user can initiate a transfer between two locations from portal running on a third system. 2 -1. 64

Grid. FTP third party transfers Fig 2. 5 2 -1. 65

Reliable. File. Transfer (RFT) service Grid. FTP is not a Web/Grid service. Reliable. File. Transfer (RFT) service provides service interface and additional features for reliable file transfers (retry capabilities etc. ). RFT uses Grid. FTP servers to effect actual transfer. 2 -1. 66

Globus Open Source Grid Software G T 4 G T 3 G T 2 G T 3 G T 4 Community Scheduler Framework [contribution] Delegation Service Python WS Core [contribution] C WS Core Community Authorization Service OGSA-DAI [Tech Preview] WS Authentication Authorization Reliable File Transfer Grid Resource Allocation Mgmt (WS GRAM) Monitoring & Discovery System (MDS 4) Java WS Core Grid. FTP Grid Resource Allocation Mgmt (Pre-WS GRAM) Monitoring & Discovery System (MDS 2) C Common Libraries Pre-WS Authentication Authorization RFT Web Services Components Non-WS Components Replica Location Service XIO Credential Management Security Data Management Execution Management Information Services Common Runtime I Foster

Globus file transfer command globus-url-copy Example Source URL globus-url-copy gsiftp: //www. coit-grid 02. uncc. edu/~abw/hello file: ///home/abw/ Destination URL copies file hello from coit-grid 02. uncc. edu to the local machine using Grid. FTP. Users needs valid security credentials (a certificate and proxy) 2 -1. 68

Question Why three /’s in file URL, i. e. file: /// ? Answer The general form of file URL is file: //host/path. If host omitted, it is assumed to be localhost, left with three /’s, i. e. file: ///. 2 -1. 69

File Staging Moving complete files to where they are needed. Usually associated with input and output files. Input file need to be moved to where program located Output files generated need to be moved back to user, or as input to other programs. Note different to input and output streaming, which moving a series of data items as a stream as it happens. 2 -1. 70

File staging Fig 2. 6 2 -1. 71

Staging example in JDD <job> … <file. Stage. Out> <transfer> <source. Url>file: ///prog 1 Out</source. Url> <destination. Url>gsiftp: //coit-grid 05. uncc. edu: 2811 /prog 1 Out</destination. Url> </transfer> </file. Stage. Out> … </job> 2 -1. 72

Staging example in JSDL <jsdl: Data. Staging> <jsdl: File. Name> /inputfiles/prog 1 Input </jsdl: File. Name> <jsdl: Creation. Flag>overwrite</jsdl: Creation. Flag> <jsdl: Source> <jsdl: URI> gsiftp: //coit grid 05. uncc. edu: 2811/prog 1 Input </jsdl: URI> </jsdl: Source> </jsdl: Data. Staging> 2 -1. 73

Sources of GT 4 information http: //www. globus. org/toolkit/docs 2 -1. 74

Questions (multiple choice) 2 -1. 75

When one issues the GT 4. 0 command: globusrun-ws -submit -F localhost: 8440 -s -so hello 1 -c /bin/echo hello what is hello? (a) A java class (b) An xml file containing the description of the job to be run (c) The executable to run in Globus (d) The argument for the program that will be executable 2 -1. 76

When one issues the GT 4. 0 command: globusrun-ws -submit -F localhost: 8440 -s -so hello 1 -c /bin/echo hello is the order of the flags important, and if so why? (a) Not important (b) Important: -c must be last as it uses the remaining arguments (c) Important: -s must be before -so (d) Important: -F must be first 2 -1. 77

When one issues the GT 4. 0 command: globusrun-ws -submit -F localhost: 8440 -s -so hello 1 -c /bin/echo hello what is localhost? (a) The server logged into running globusrun-ws. (b) The computer you are using to log into the server (c) None of the other answers. 2 -1. 78

What does the tag <count> specify in an RSL-2/JDD file? (a) The number of different jobs submitted. (b) The number of computers to use. (c) The number of identical jobs to submit. (d) The number of arguments. 2 -1. 79