Main Sphinx Design Concepts

1
Main Sphinx Design Concepts

• There are two primary design components which
  comprise Sphinx
• The Database Warehouse
• The Control Process
• The Database Warehouse is the information depot
  for all of Sphinx
• All information is stored/retrieved via the
  database (or possibly multiple databases)
• There is no direct communication between
  different internal Sphinx modules
• The API for all internal modules is SQL along
  (with the database schema)
• The Control Process is the main point of activity
• It is a loop which calls different Sphinx modules
• All modules in the control process operate on
  stateful entities in the database, changing the
  state of an entity from one to another state
• Sphinx operates as a Finite State Machine!
• Uses data mining principles in the design
• The API for all modules is SQL along the database
  schema
• Modules mine the database for information and
  then take some action.

2
Version 1 Sphinx Server/Client
Sphinx Server
VDT Client
Sphinx Client
Data Warehouse
Message Interface
Message Interface
DAG Reducer
Single-threaded Control Process
Job Tracker
Prediction Engine
Job Submitter
Site Tables
Condor-G/DAGMan
Job Execution Planner
VDT Server Site
Resource
Job Table
DAG Table
Replica Table
Information Gatherer
Monitoring Service
Message Tbl.
3
More Details

• The next slides show the message passing between
  different Sphinx components
• The components are split (packaged) in the
  following way
• Sphinx Client
• Incomming Messages
• Outgoing Messages
• Sphinx Server
• Incomming Messages
• Control Process
• Outgoing Messages
• See http//ufgrid.phys.ufl.edu8080 for
  documentation
• The next version of Sphinx (currently in
  progress) will split/package the components into
  a more logically distinct, modular scheme (yet to
  be decided)

4
Version 1 Sphinx Client (outgoing message)
Message ID Msg Type Msg Sender Message
XML RPC
XML message Destination URL
Clinet
ClinetOutModule
ACK
ACK
ACK
Files
MsgID Msg Type Msg Sender Message
Message Type Path to file(DAX) Path to User ID
file
XML message
Msg strings Sender Info
XMLDocUtils
MsgID Msg Type Msg Sender Message
XML message
5
Version 1 Sphinx Client (incoming message)
JobWrapper
Message string
job ID dag ID
SubmissionGenerator
Msg string Work Dir
ACK
Super dag file
ClientIncModule
createSubmission
execute
XML message
Msg type Message string
condor_submit_dag
XML message
MsgHandler
Msg ID Msg Type Msg Sender Message
XML message
6
Version 1 Sphinx Server (incoming message)
DB Connector Msg Type Msg Sender Info Message
XML message
MessageWrapper
InMessageWrapper
IncModule
ACK (Msg ID)
Msg Id Msg Type Msg Sender Info Message
checkIn
XML message
Dag
MsgHandler
message
Msg Id Msg Type Msg Sender Info Message
checkIn
XML message
checkOut
remove(msg)
Job
DOMparser
AbstractDagParser
checkIn
Dag Obj
Job Obj
Dag Info Job Info
Dagwrapper
Jobwrapper
7
Version 1 Control Process
Control Process
UNREDUCED
DagReducer
Dag Table
UNPREDICTED
UNPREDICTED
PredictionEngine
UNFINISHED
UNPREDICTED
Planner
SchedulingModule
UNPLANNED
Job, Strategy
execute
Job Table
Sites Info
UNPLANNED
Exec. site info Input file info Output file info
UNSENT
Physical site info
Site Tables
JobWrapper
Client ID Msg tpye Job info string
Physical file info
Message Table
Check In / SUB_REQ
Replica Table
MessageWrapper
OutMessageWrapper
8
Version 1 Sphinx Server (outgoing message)
Msg ID Client Info Msg type message
XML message Destination URL
SOAP RPC
OutgoingInterface
SendRequest
ACK
ACK
ACK
MsgID Msg Type Client Info Message
Next msg ID
message
XML message
MessageWrapper
Check Out
MsgID Msg Type Client Info Message
XML message
Message
9
Version 2 of Sphinx (current)

• The current development cycle is represented in
  the next three slides
• Essentially
• More planning tables have been added to the
  database warehouse
• More planning modules have been added to the
  control process
• A Data Management Component is being added to the
  Sphinx Server
• Replica Management (via RLS and GridFTP)
• The Monitoring Interface was upgraded to use
  MonALISA
• We also want to investigate the possibility of
  making a Clarens service out of any arbitrary
  planning module in the control process
• The Job Submitter, Job Tracking, and
  Condor-G/DAGMan (which currently form the Sphinx
  Client) is an example.

10
Version 2Sphinx Server with Sphinx Submission
Client
Sphinx Server
VDT Client
Sphinx Client
Data Warehouse
Message Interface
Message Interface
DAG Reducer
Multi-threaded Control Process
Job Tracker
Job Predictor
DAG Predictor
Policies
Job Submitter
Accounting
Job Admission Control
Resource Prop.
Condor-G/DAGMan
DAG Admission Control
Resource Hier.
DAG Data Repl. Planner
Resource Dist.
Distance
Job Execution Planner
Entity-Name
VDT Server Site
Entity Hier.
Resource
Job-Entity Prop.
Data Management
Job
DAG Tracker
DAG
Rel. Data
Repl. Loc. Serv.
Rel. Appl.
Information Gatherer
Job Tracking
Grid Weather
Monitoring Service
Schedule
Prediction
Message
11
Version 2Sphinx Server with Sphinx UI Client
Sphinx Server/VDT Client
UI Machine
Sphinx Client
Data Warehouse
Message Interface
Message Interface
DAG Reducer
Multi-threaded Control Process
Job Predictor
DAG Predictor
Policies
Accounting
Job Admission Control
Resource Prop.
DAG Admission Control
Resource Hier.
DAG Data Repl. Planner
Resource Dist.
Distance
Job Execution Planner
Entity-Name
Condor-G/DAGMan
VDT Server Site
Job Submitter
Entity Hier.
Resource
Job Tracker
Job-Entity Prop.
Data Management
Job
DAG Tracker
DAG
Rel. Data
Repl. Loc. Serv.
Rel. Appl.
Information Gatherer
Job Tracking
Grid Weather
Monitoring Service
Schedule
Prediction
Message
12
Version 2 Finite State Machine Control Process
DAG unreduced JOB unpredicted
Await Mesg.
no
yes
DAG unpredicted JOB unpredicted
DAG unfinished JOB unplanned
Reduce DAG
Plan Dag
no
no
yes
yes
DAG unpredicted JOB unaccepted
DAG unfinished JOB unsent
Predict Job
Plan Job
no
no
yes
yes
DAG unaccepted JOB unaccepted
DAG unfinished JOB remove
Predict DAG
Submit Job
no
no
yes
yes
DAG unaccepted JOB unplanned
DAG remove JOB remove
Admit Job
Remove Job
no
no
yes
yes
DAG unplanned JOB unplanned
Remove DAG
Admit DAG
no
yes