maDagAgent man page

maDagAgent — DIET grid middleware agent

Name

maDagAgent - Main binary of the DIET Architecture for managing workflows

Synopsys

maDagAgent config_file [sched] [pfm] [IRD] ...

Description

The maDagAgent is the daemon responsible for the management of workflows execution.

Diet Platform

A DIET platform is buildt upon Server Daemons (SeD). Requests are distributed amongst a hierarchy of agents. The scheduler can rely on resources availability informations collected from three different tools: NWS sensors which are placed on every node of the hierarchy, from the application-centric performance prediction tool FAST which relies on NWS informations or from CoRI Easy a module based on simple system calls and basic performance tests.

The different components of a DIET architecture are the following:

Client

A client is an application which uses DIET to solve computational problems. Clients could be web pages, PSE scripts like Matlab or Scilab or native program.

Master Agent (MA)

A MA manages computation requests from clients. It chooses the best server available to handle the request based on performance informations collected from servers. Then, the reference of the chosen server is returned to the client.

Local Agent (LA)

A LA transmits requests between MAs and servers. LAs store a list of services available in their subtree. For each service, LAs store a list of children (either agents or servers) providing the former. Depending on the underlying network topology, a hierarchy of LAs may exists between the MA and the appropriate servers, one of LAs tasks is to do a partial scheduling on its subtree, effectively reducing its MA workload.

Server Daemon (SeD)

A SeD encapsulate a computational resource. FOr instance, it can be locate on the entry point of a parallel computer. SeD store a list of locally available data, available computational solvers and performance-related information (available memory amount or number or resources). During registration, SeD declare to its parent agent (LA or MA) every computational problem it can solve. SeD can send performance and hardware informations by using the CoRI module or performance predictions for some kinds of problems by using the FAST module.

Master Agent DAG (MA DAG)

The Master Agent DAG (MADAG) provides DAG workflow scheduling. This agent serves as the entry point to the Diet Hierarchy for a client that wants to submit a workflow. The language supported by the MADAG is based on XML.

Corba Usage for Diet

DIET relies on the CORBA naming service for service discovery allowing every entity to interconnect. Reference to the omniORB naming service is written down in a CORBA configuration file whose path is given to omniORB through the environment variable OMNIORB_CONFIG.

The  lines  concerning  the name server in the omniORB configuration file are built as follows:

InitRef  =  NameService=corbaname::<hostname>:<port>

The  name  server port is the port given as an argument to the -start option of omniNames. You also need to update your LD LIBRARY PATH to  point to <install dir>/lib. So your LD LIBRARY PATH environment variable should now be :

LD LIBRARY PATH=<omniORB home>/lib:<install dir>/lib.

NB1:  In order to avoid name collision, every agent must be  assigned  a different name in the name server; since they don't have any children, SeDs do not need names assigned to them and they don't register with the name server.

NB2:  Each  Diet  hierarchy can use a different name server, or multiple hierarchies can share one name server (assuming all agents are assigned unique names).  In  a multi-MA environment, in order for multiple hierarchies to be able to cooperate it is necessary that they all share the same name server.

Options

config_file

Configuration file used by the agent to launch the DIET entity

sched

The policy used for scheduling workflows. This option can take the following values:

  • -basic (default):
  • -g_heft:
  • -g_aging_heft:
  • -fairness:
  • -srpt:
  • -fcfs: first come first serve
pfm
  • -pfm_any (default)
  • -pfm_sameservices
IRD
  • -IRD value

Diet Configuration File

Every DIET entity requires a configuration file.

Please note that:

Depending on the type of DIET element, different kinds of  keyword  could be  found.  Here is a list of the possible keywords for a DIET Agent configuration file:

traceLevel

Integer value corresponding to the traceLevel for the DIET agent:

  • 0: DIET do not print anything.
  • 1: DIET prints only warnings and errors on the standard error output.
  • 2: [default] DIET prints information on the main steps of a call.
  • 5: DIET prints information on all internal steps too.
  • 10: DIET prints all the communication structures too.
  • >10: (traceLevel - 10) is given to the ORB to print CORBA messages too.

0 Diet do not print anything,
1 Diet prints only warnings and errors on the standard error output,
2 [default] Diet prints information on the main steps of a call,
5 Diet prints information on all internal steps too,
10 Diet prints all the communication structures too,
> 10 (traceLevel - 10) is given to the ORB to print CORBA messages too.

agentType

three possible values

  • DIET MASTER AGENT (or MA) for a Master Agent
  • DIET LOCAL AGENT (or LA) for a Local Agent
  • DIET_MA_DAG for an MA DAG Agent
dietPort

Integer setting the listening port of the agent. If left empty, the ORB will get an open port from the system (if default 2809 is busy).

dietHostName

String setting the listening interface of the agent. If left empty, the ORB will use the system hostname (the first one if several are available).

name

String identifying the element. Clients and children nodes (LAs and SeDs) must point to the same CORBA Naming Service hosting the MA.

parentName

String identifying the parent agent.

[Remark: Only DIET Local Agents could use the parentName keyword]

fastUse

Boolean enabling/disabling FAST module. If set to 0, all LDAP and NWS parameters are ignored, and all requests to FAST are disabled (when Diet is compiled with FAST). This is useful while testing a DIET platform without having to deploy an LDAP base nor an NWS platform.

[Remark: DIET must be compiled with FAST]

ldapUse

Boolean enabling/disabling LDAP support.

[Remark: DIET must be compiled with FAST]

ldapBase

String representing the LDAP base storing FAST-known services address in the form host:port .

[Remark: DIET must be compiled with FAST]

ldapMask

String specifying the mask registered in the LDAP base.

[Remark: DIET must be compiled with FAST]

nwsUse

Boolean enabling/disabling NWS support.

[Remark: DIET must be compiled with FAST]

nwsNameserver

String representing the NWS naming service address in the form host:port .

[Remark: DIET must be compiled with FAST]

nwsForecaster

String representing the NWS forecast module used by FAST.

[Remark: DIET must be compiled with FAST]

useLogService

Boolean enabling/disabling the LogService for monitoring purposes.

lsOutbuffersize

Integer setting outgoing messages buffer size.

lsFlushinterval

Integer setting the flush interval for the outgoing messages buffer.

neighbours

String listing MA that must be conntacted to build a federation. It is formatted as a white-space separated list of addresses in the form host:port.

[Remark: DIET must be compiled with the Multi-MA option]

minimumNeighbours

Integer setting the minimum connected neighbours. If the agent has less connected neighbours, it will try establishing new connections.

[Remark: DIET must be compiled with the Multi-MA option]

maximumNeighbours

Integer setting the maximum connected neighbours. Further, the agent will refuse newer connections.

[Remark: DIET must be compiled with the Multi-MA option]

updateLinkPeriod

Integer setting the period (in seconds) at which the agent will check its neighbours status and will try establishing new connections if their numbers is less than minimumNeighbours.

[Remark: DIET must be compiled with the Multi-MA option]

bindServicePort

Integer defining the port used by the MA to share its IOR.

[Remark: Option used only by MAs]

useConcJobLimit

Boolean enabling/disabling the SeD restriction about concurrent solves. This should be used in conjunction with maxConcJobs.

[Remark: Option used only by SeDs]

maxConcJobs

Integer setting the maximum number of jobs running at once. This should be used in conjunction with maxConcJobs.

[Remark: Option used only by SeDs]

locationID

String used for alternative transfer cost prediction in Dagda.

[Remark: Option used only by SeDs]

MADAGNAME

String corresponding to the name of the MADAG agent.

[Remark: DIET must be compiled with the workflow option]

[Remark: Option used only by clients]

schedulerModule

Path to the sheduler library module containing the scheduler implementation.

[Remark:  DIET  must  be compiled with the User Scheduling option]

[Remark: Option used only by agents]

moduleConfigFile

String corresponding to an optional configuration file for the module.

[Remark: DIET must be compiled with the User Scheduling option]

[Remark: Option used only by agents]

batchName

String corresponding to the name of the queue where the job will be submitted.

[Remark: DIET must be compiled with the Batch option]

[Remark: Option used only by SeDs]

pathToNFS

Path to the NFS directory where you have read/write permissions.

[Remark: DIET must be compiled with the Batch option]

[Remark: Option used only by SeDs]

pathToTmp

Path to the temporary directory where you have read/write permissions.

[Remark: DIET must be compiled with the Batch option]

[Remark: Option used only by SeDs]

internOARbatchQueueName

String only useful when using CORI batch features with OAR 1.6

[Remark: DIET must be compiled with the Batch option]

[Remark: Option used only by SeDs]

initRequestID

Integer setting the MA Request ID starting value.

[Remark: Option used only by MAs]

maxMsgSize

Integer setting the maximum size of CORBA messages sent by Dagda. By default, it's the same as the omniORB giopMaxMsgSize size.

maxDiskSpace

Integer setting maximum disk space available to Dagda for storingt data. When set to 0, Dagda will ignore any disk quota. By default, it's the same value as available disk space on the partition set by storageDirectory.

maxMemSpace

Integer setting the maximum memory available to Dagda. When set to 0, Dagda will ignore any memory usage limitation. By default, no limitations.

cacheAlgorithm

String defining the cache replacement algorithm used when Dagda needs more memory for storing a piece of data. Possible values are: LRU, LFU, FIFO. By default, no cache replacement algorithm, Dagda never overwrite data.

shareFiles

Boolean enabling/disabling Dagda file sharing with its children. Requires that the path is accessible by the children (ie: NFS partition shared by parent and children). By default, no file sharing.

dataBackupFile

Path to the backup file used by Dagda on user request (checkpointing). By default, no checkpointing is disabled.

[Remark: Option used by Agents and ServerDaemon]

restoreOnStart

Boolean defining if Dagda have to load the file set by dataBackupFile at startup and restore all data recorded during the last checkpointing event. Disabled by default.

[Remark: Option used by agents and SeDs]

storageDirectory

String defining the directory where Dagda will store data files. By default /tmp is used.

Environment

DIET needs some variables to be defined in order for the agent to be able to find all the mandatory library and the CORBA naming service.

LD_LIBRARY_PATH

This environment variable must contain the path to the omniORB libraries

OMNIORB CONFIG

Path to the CORBA configuration file where the reference to the omniORB naming service is written.

Examples

Here are examples of configuration file for the MA DAG Agent.

traceLevel = 2
agentType = DIET_MA_DAG
name = mad
parentName = MA1

Reporting Bugs

If you find that software interesting, or if you find a bug, please send us a mail: <diet-dev@ens-lyon.fr> with the description of the problem, the version of the program and/or any information that could help us fixing it.

Authors

GRAAL
INRIA Rhone-Alpes
46 allee d'Italie 69364 Lyon cedex 07, FRANCE
Email: <diet-dev@ens-lyon.fr>
www: http://graal.ens-lyon.fr/DIET

SysFera
13 avenue Albert Einstein
69100 Villeurbanne, FRANCE
Email: <contact@sysfera.com>
www: http://www.sysfera.com

See Also

omniNames(1), dietAgent(1), dietForwarder(1)

Author

benjamin.depardon@sysfera.com

License: CeCILL

Referenced By

dietAgent(1).

2011-05-23 0.1