Supercharging your Apache OODT deployments with the

Process Control SystemChris A. Mattmann

NASA JPL/Univ. Southern California/ASFmattmann@apache.org November 9, 2011

• Apache Member involved in– OODT (VP, PMC), Tika (VP,PMC), Nutch (PMC), Incubator (PMC), SIS

(Mentor), Lucy (Mentor) and Gora (Champion), MRUnit (Mentor), Airavata (Mentor)

• Senior Computer Scientist at NASA JPL in Pasadena, CA USA

• Software Architecture/Engineering Prof at Univ. of Southern California

And you are?

Welcome to the Apache in Space! (OODT) Track

Agenda

• Overview of OODT and its history• What is the Process Control System (PCS)?• PCS Architecture• Some hands on examples

– Health Monitoring– Pedigree/Provenance

• Deploying PCS • Where we’re headed

• Increasing data volumes (exponential growth)

• Increasing complexity of instruments and algorithms

• Increasing availability of proxy/sim/ancillary data

• Increasing rate of technology refresh

… all of this while NASA Earth Mission funding was decreasing

A data system framework based on a standard architecture and reusable software components for supporting all future missions.

Lessons from 90’s era missions

Object Oriented Data Technology http://oodt.apache.org

Funded initially in 1998 by NASA’s Office of Space Science

Envisaged as a national software framework for sharingdata across heterogeneous, distributed data repositories

OODT is both an architecture and a reference implementation providing

Data ProductionData Distribution

Data DiscoveryData Access

OODT is Open Source and available from the Apache Software Foundation

Enter OODT

Apache OODT• Originally funded by NASA to focus on

– distributed science data system environments

– science data generation – data capture, end-to-end– Distributed access to science data

repositories by the community• A set of building blocks/services to exploit

common system patterns for reuse • Supports deployment based on a rich

information model• Selected as a top level Apache Software

Foundation project in January 2011• Runner up for NASA Software of the Year• Used for a number of science data

system activities in planetary, earth, biomedicine, astrophysics

http://oodt.apache.org

Apache OODT Press

Why Apache and OODT?• OODT is meant to be a set of tools to

help build data systems– It’s not meant to be “turn key” – It attempts to exploit the boundary

between bringing in capability vs. being overly rigid in science

– Each discipline/project extends

• Apache is the elite open source community for software developers– Less than 100 projects have been

promoted to top level (Apache Web Server, Tomcat, Solr, Hadoop)

– Differs from other open source communities; it provides a governance and management structure

Apache OODT Community

• Includes PMC members from– NASA JPL, Univ. of Southern California, Google, Children’s

Hospital Los Angeles (CHLA), Vdio, South African SKA Project

• Projects that are deploying it operationally at– Decadal-survey recommended NASA Earth science

missions, NIH, and NCI, CHLA, USC, South African SKA project

• Use in the classroom– My graduate-level software architecture and seach engines

courses

OODT Framework and PCS

OBJECT ORIENTED DATA TECHNOLOGY FRAMEWORK

OODT/Science Web Tools

ArchiveClient

ProfileXML Data

DataSystem 1

DataSystem 2

ArchiveService

ProfileService

ProductService

QueryService

Bridge to ExternalServices

Navigation Service

OtherService 1

OtherService 2

Catalog & ArchiveService

Process Control System (PCS)

Catalog & ArchiveService (CAS)

CAS has recently become known as Process Control Systemwhen applied to mission work.

Orbiting Carbon Observatory (OCO-2) - spectrometer instrumentNASA ESSP Mission, launch date: TBD 2013PCS supporting Thermal Vacuum Tests, Ground-based instrument data processing, Space-based instrument data processing and Science Computing FacilityEOM Data Volume: 61-81 TB in 3 yrs Processing Throughput: 200-300 jobs/day

NPP Sounder PEATE - infrared sounderJoint NASA/NPOESS mission, launch date: October 2011

PCS supporting Science Computing Facility (PEATE)EOM Data Volume: 600 TB in 5 yrs Processing Throughput: 600 jobs/day

QuikSCAT - scatterometerNASA Quick-Recovery Mission, launch date: June 1999

PCS supporting instrument data processing and science analyst sandbox

Originally planned as a 2-year mission

SMAP - high-res radar and radiometerNASA decadal study mission, launch date: 2014

PCS supporting radar instrument and science algorithm development testbed

Current PCS deployments

Other PCS applications

BioinformaticsNational Institutes of Health (NIH) National Cancer Institute’s (NCI) Early Detection Research Network (EDRN)

Children’s Hospital LA Virtual Pediatric Intensive Care Unit (VPICU)

Technology DemonstrationJPL’s Active Mirror Telescope (AMT)

White Sands Missile Range

Earth ScienceNASA’s Virtual Oceanographic Data Center (VODC)

JPL’s Climate Data eXchange (CDX)

Astronomy and RadioPrototype work on MeerKAT with South Africans and KAT-7 telescope

Discussions ongoing with NRAO Socorro (EVLA and ALMA)

• All Core components implemented as web services– XML-RPC used to communicate between components– Servers implemented in Java– Clients implemented in Java, scripts, Python, PHP and web-apps– Service configuration implemented in ASCII and XML files

PCS Core Components

• File Manager does Data Management– Tracks all of the stored data, files & metadata– Moves data to appropriate locations before and after initiating PGE runs and from staging area to

controlled access storage

• Workflow Manager does Pipeline Processing– Automates processing when all run conditions are ready– Monitors and logs processing status

• Resource Manager does Resource Management– Allocates processing jobs to computing resources– Monitors and logs job & resource status– Copies output data to storage locations where space is available– Provides the means to monitor resource usage

Core Capabilities

PCS Ingestion Use Case

File/Metadata Capabilities

PCS Processing Use Case

Advanced Workflow Monitoring

Resource Monitoring

PCS Support for OCOOCO has three PCS deployments (installation of core components):

– Thermal Vacuum Instrument Testing deployment– A PCS configuration was successfully deployed to process and analyze 100% of all L1a products

generated during t-vac testing – Space-based Operations & Ground-based FTS processing deployment

– Automatic processing of all raw instrument data through AOPD L2 algorithm– Currently operational, our FTS deployment has processed over 4 TB of FTS spectrum and FTS L1a

products for science analysis to data

– Science Computing Facility (SCF) deployment– Supports all L2 full physics algorithm processing for science analysis and cal/val– Supports scientists’ investigations of alternative algorithms & data products

Ability to adapt to change– Scaled up the database catalog size

– When size grew > 1 million products, moved from Lucene to Oracle in a weekend!

– Had to repartition the FTS archive layout and structure 2 years into the mission– Recataloged all 1 million FTS products and moved all data within a few weeks!

– Accommodated Ops/SCF hardware reconfiguration 1 year prior to launch– Physically-shared and virtually-separated to virtually-shared and physically-separated at no extra cost!

OCO Hardware Environment with PCS

Source: S. Neely, OCO Hardware Review

How do we deploy PCS for a mission?• We implement the following mission-specific customizations

– Server Configuration• Implemented in ASCII properties files

– Product metadata specification• Implemented in XML policy files

– Processing Rules• Implemented as Java classes and/or XML policy files

– PGE Configuration• Implemented in XML policy files

– Compute Node Usage Policies• Implemented in XML policy files

• Here’s what we don’t change– All PCS Servers (e.g. File Manager, Workflow Manager, Resource Manager)

• Core data management, pipeline process management and job scheduling/submission capabilities

– File Catalog schema

– Workflow Model Repository Schema

Server and PGE Configuration

What is the Level of Effort for personalizing PCS?

• PCS Server Configuration – “days”– Deployment specific

• Addition of New File (Product) Type – “days”– Product metadata specification– Metadata extraction (if applicable)– Ingest Policy specification (if remote pull or remote push)

• Addition of a New PGE – (initial integration, ~ weeks)– Policy specification– Production rules– PGE Initiation

* Estimates based on OCO and NPP experience

A typical PCS service (e.g., fm, wm, rm)

What’s PCS configuration?• Configuration follows typical Apache-like server

configuration– A set of properties and flags that are set in an ASCII text file that

initialize the service at runtime

• Properties configure– The underlying subsystems of the PCS service

• For file manager, properties configure e.g., – Data transfer chunk size– Whether or not the catalog database should use quoted strings for

columns– What subsystems are actually chosen (e.g, database versus Lucene,

remote versus local data transfer)

• Can we see an example?

PCS File Manager Configuration File

•Set runtime properties•Choose extension points•Sensible defaults if you don’t want to change them

What’s PCS policy?• Policy is the convention in which missions define

– The products that should be ingested and managed– The PGE default input parameters and their required data and

metadata inputs (data flow)– The PGE pre-conditions and execution sequence (control flow)– The underlying hardware/resource environment in which PGEs

should run and data/metadata should be captured• What nodes are available?• How much disk space is available?• How should we allocate PGEs to nodes?

• Can we see an example?

PCS File Manager Policy for OCO FTS Products

•The scheme for laying out products in the archive•The scheme for extracting metadata from products on the server side•A name, ID an description of the each product

PCS Workflow Manager Policy for OCO FTS Products

•Define data flow•Define control flow

PCS Overall Architecture

•What have we told you about so far?

•What are we going to tell you about now?

The concept of “production rules”• Production rules are common terminology to refer

to the identification of the mission specific variation points in– PGE pipeline processing– Product cataloging and archiving

• So far, we’ve discussed– Configuration – Policy

• Policy is one piece of the puzzle in production rules

Production rule areas of concerns1. Policy defining file ingestion

1. What metadata should PCS capture per product?2. Where do product files go?

2. Policy defining PGE data flow and control flow3. PGE pre-conditions 4. File staging rules5. Queries to the PCS file manager service

1-5 are implemented in PCS (depending on complexity) as either:1. Java Code2. XML files3. Some combination of Java code and XML files

PCS Task Wrapper aka CAS-PGE

•Gathers information from the file manager

•Files to stage•Input metadata (time ranges, flags, etc.)

•Builds input file(s) for the PGE•Executes the PGE•Invokes PCS crawler to ingest output product and metadata•Notifies Workflow and Resource Managers about task (job) status•Can optionally

•Generate PCS metadata files

PCS experience on recent missions• How long did it take to build out the PCS configuration and policy?

– For OCO, once each pipeline was system engineered and PGEs were designed

• Level of Effort– Configuration for file, resource, workflow manager : 1-2 days (1x cost)– Policy for file, resource, workflow manager: 1-2 days per new PGE and new Product Type– Production rules: 1 week per PGE

– For NPP Sounder PEATE, once each PGE was system engineered and designed

• Level of Effort– Configuration for file, resource, workflow manager : 1-2 days (1x cost)– Policy for file, resource, workflow manager: 1-2 days per new PGE and new Product Type– Production rules: 1 week per PGE

• Total Level of Effort– OCO: 1.0 FTEs over 5 years– NPP Sounder PEATE: 2.5 FTEs over 3 years

Some relevant experience with NRAO: EVLA prototype

• Explore JPL data system expertise– Leverage Apache OODT– Leverage architecture experience– Build on NRAO Socorro F2F given in April 2011 and

Innovations in Data-Intensive Astronomy meeting in May 2011

• Define achievable prototype– Focus on EVLA summer school pipeline

• Heavy focus on CASApy, simple pipelining, metadata extraction, archiving of directory-based products

• Ideal for OODT system

Architecture

Pre-Requisites

• Apache OODT– Version: 0.3-SNAPSHOT

• JDK6, Maven2.2.1• Stock Linux box

Installed Services• File Manager

– http://ska-dc.jpl.nasa.gov:9000• Crawler

– http://ska-dc.jpl.na.gov:9020• Tomcat5

– Curator: http://ska-dc.jpl.nasa.gov:8080/curator/– Browser: http://ska-dc.jpl.nasa.gov/ – PCS Services: http://ska-dc.jpl.nasa.gov:8080/pcs/services/ – CAS Product Services: http://ska-dc.jpl.nasa.gov:8080/fmprod/ – Workflow Monitor: http://ska-dc.jpl.nasa.gov:8080/wmonitor/

• Met Extractors– /usr/local/ska-dc/pge/extractors (Cube, Cal Tables)

• PCS package– /usr/local/ska-dc/pcs (scripts dir contains pcs_stat, pcs_trace, etc.)

Demonstration Use Case• Run EVLA Spectral Line Cube generation

– First step is ingest EVLARawDataOutput from Joe– Then fire off evlascube event– Workflow manager writes CASApy script dynamically

• Via CAS-PGE– CAS-PGE starts CASApy– CASApy generates Cal tables and 2 Spectral Line Cube

Images– CAS-PGE ingests them into the File Manager

• Gravy: UIs,Cmd Line Tools, Services

Results: Workflow Monitor

Results: Data Portal

Results: Prod Browser

Results: PCS Trace Cmd Line

Results: PCS Stat Cmd Line

Results: PCS REST Services: Trace

curl http://host/pcs/services/pedigree/report/flux_redo.cal

Results: PCS REST Service: Health

curl http://host/pcs/services/health/report

Read up on https://issues.apache.org/jira/browse/OODT-139Read documentation on PCS services: https://cwiki.apache.org/confluence/display/OODT/OODT+REST+Services

Results: RSS feed of prods

Results: RDF of products

Where are we headed?

• OPSui work– OODT-157 you will have heard about this earlier in

the day from Andrew Hart• Improved PCS services

– Integrate more services into OODT-139 including curation services, and workflow services for processing

• Workflow2 improvements described in OODT-215

Where are we headed

• Integration with Hadoop Nextgen M/R– http://svn.apache.org/repos/asf/oodt/

branches/wengine-branch/ • Integration with more catalogs

– Apache Gora, MongoDB• Integration with GIS services

– GDAL, regridding, etc.• Improved science algorithm wrapping

OODT Project Contact Info• Learn more and track our progress at:

– http://oodt.apache.org – WIKI: https://cwiki.apache.org/OODT/ – JIRA: https://issues.apache.org/jira/browse/OODT

• Join the mailing list:– dev@oodt.apache.org

• Chat on IRC:– #oodt on irc.freenode.net

• Acknowledgements– Key Members of the OODT teams: Chris Mattmann, Daniel J. Crichton, Steve Hughes,

Andrew Hart, Sean Kelly, Sean Hardman, Paul Ramirez, David Woollard, Brian Foster, Dana Freeborn, Emily Law, Mike Cayanan, Luca Cinquini, Heather Kincaid

– Projects, Sponsors, Collaborators: Planetary Data System, Early Detection Research Network, Climate Data Exchange, Virtual Pediatric Intensive Care Unit, NASA SMAP Mission, NASA OCO-2 Mission, NASA NPP Sounder Peate, NASA ACOS Mission, Earth System Grid Federation

Alright, I’ll shut up now

• Any questions?

• THANK YOU!– chris.a.mattmann@nasa.gov – @chrismattmann on Twitter