Transcript PSPS-CDR-day2

Pan-STARRS PS1
Published Science Products Subsystem
Critical Design Review
November 5-6, 2007
Honolulu
CDR – Day 2
slide 2
Topics: November 6 / Day 2
•
•
•
•
•
•
•
•
•
•
•
Welcome back (Heasley)
ODM Continued (JHU team)
The PSPS Data Retrieval Layer (Referentia team)
Lunch
The Web Based Interface (Heasley)
Pan-STARRS External Interfaces (Heasley)
PSPS Test Plan (Heasley)
PSPS Schedule (Heasley)
System Level Risk Assessment (Heasley)
Executive Session (Committee only)
Recap Meeting of Review Panel with Subsystem and component
leads
• Adjourn
slide 3
The Object Data Manger System
Continued
The Johns Hopkins Team
slide 4
The PSPS Data Retrieval Layer
The Referentia Team
slide 5
Pan-STARRS PS1
Published Science Products Subsystem
Critical Design Review
Data Retrieval Layer (DRL)
Referentia Systems, Inc
Matt Shawver, [email protected], 808-423-1900x111
Kenn Yuen, [email protected]
Chris Richmond, [email protected]
Jay Knight, [email protected]
Nov 5-6, 2007
slide 6
Outline







slide 7
Software Architecture
Requirements and Implementation Plans
Key Design Modifications
Test Plan
Development Schedule
DRL Development Status
Demo
High Level Software Architecture
slide 8
DRL Software Architecture
Web Based
Interface
Test Web Based
Interface
Generic Web
Service Client
Java Web Service Proxy
Web Service Interface
Result Set Persistence
(Java Caching System)
DM Adaptor
Query
Manager
Adaptor
PostgreSQL
JDBC
Adaptor
MySQL
JDBC
Adaptor
JHU Query
Manager
PostgreSQL
Driver
MySQL
Driver
Data Managers
slide 9
Login / Session Manager
Tomcat Security
DRL Requirements





Query Analysis
Query Queuing and Execution
Result Caching
Result Retrieval
Administrative
• Performance Monitoring
• User Administration
• Logging
 Support multiple Data Managers
• JHU Query Manager
• MySQL
• SQL Server
• PostgreSQL
slide 10
Req: Query Analysis
From previous design requirements:
 Syntax validation
 Current DM Resource Load
 Query processing time estimate
 Schema information
slide 11
Query Analysis Implementation
 For syntax validation, SQL Server PARSEONLY
command
 For performance status, SQL Server sp_monitor
procedure
 Highly database implementation dependent
 Exact prediction of query time is impossible. Instead
show query execution plan.
 Schema information will be retrieved by querying
database metadata views and functions
 For ODM, use Query Manager functionality when
available
slide 12
Req: Query Queuing and
Execution
From previous design requirements:
 Query any of the Data Managers
 Provide status and progress information
 Provide priority level with validation status results
 Set query priority based on validation priority level
slide 13
Query Queuing and Execution
Implementation
 Issue: database query execution plans are not always accurate.
 Alternative implementation: treat all queries the same at first.
 Short, medium and long queues each have their own connections
allocated
 If a short or medium query takes longer than a certain amount of
time, it will be moved to a longer queue
 Queue sizes, and expiration times will be user configurable
 If the long queue runs out of query slots, most recent query will be
cancelled and restarted when a slot becomes available.
 For ODM, use Query Manager queuing functionality (user chooses
which queue to use)
slide 14
Req: Result Caching
From previous design requirements:
Query result sets are stored in the DRL Cache
until they have been retrieved by a PDC.
Purge retrieved result sets if space is needed
slide 15
Result Caching Implementation
 Maintain results in result set cache as long as possible
to allow repeated retrieval of results
 With a large enough cache (terabyte), results should be
typically held for a week or more
 Link to past results via query history
 Performance of the result set cache is critical for PSPS
responsiveness
• Hybrid memory / disk cache
– LRU for memory and disk
• In-memory index for fast disk retrieval
• Retrieval of partial results
• Efficiently support writing and reading multiple
concurrent result sets
slide 16
Result Caching (continued)
 Java Caching System (JCS) Implementation
• Web server caching system
• Uses Java serialization
• In memory storage, with swapping to indexed file on disk
• Built-in capability for distributing cache across multiple
machines (untested)
• Modified JCS to support synchronous puts when memory is full
(wait for space to be freed via disk write)
• Store Result Set as a list of objects each made up of a block of
rows
• Support many result sets (each result set can use as little as
one block of rows in memory)
• Adding memory speeds cache
slide 17
Req: Result Retrieval
From previous design requirements:
•The PS1 DRL shall return query results in
response to a query request.
slide 18
Result Set Retrieval
Implementation
 Don’t slow down fast queries
• Return results immediately if query is very fast
 Enable incremental results for queries with large data
volumes
 Status updates with number of rows retrieved
 Execution status if supported by database
 Support streaming CSV file download
• Stream file directly from cache rather than creating
on disk
slide 19
Req: Performance Monitoring
From previous design requirements:
 Allow administrators to monitor the performance of
the DRL functions
• I/O statistics
• CPU statistics
• memory statistics
• process statistics
 at configurable levels of detail
slide 20
Performance Monitoring
Implementation
 JMX over RMI to provide management interface to all
JVM information collected at runtime
• Does not provide CPU information
 Use cross platform third party library if more detailed
information is required
• YourKit (http://www.yourkit.com) is one good third
party option
• Tradeoff: non-JVM profiling libraries incur overhead
 Provide user configurable logs to a database to store
historical information
slide 21
Req: User Administration
From previous design requirements:
 The PS1 DRL shall provide a computer security
system to protect the DRL and PSPS DM
Components from unauthorized access.
 The PS1 DRL shall provide and authenticate at least
two levels of access.
 The PS1 DRL shall provide and authenticate
privileged access for use of the private Administrative
API.
 The PS1 DRL shall provide and authenticate
standard access for use of the public API.
slide 22
User Administration
Implementation
 Initial plan: Tomcat Realm JDBC based security with an
in-process database
• Straightforward
• Independent of other components
• Allows administrator to create and modify user
accounts through web service
• Allows association of additional information with
user account
– Role
– Query log
– Available result sets
– Running queries
slide 23
Req: Logging
From previous design requirements:
 Log major system events
• Query events
• Unsuccessful Authentication Attempts
• Server restarts
• Any errors
slide 24
Logging Implementation
 Log results via JDBC to in-process database
 Move to external database if DRL is clustered in the
future
 Logs linked to user accounts (stored in same database)
slide 25
Key Modifications: DRL – DM ICD
Changes
 For non-ODM Data Managers, DRL should utilize JDBC
directly rather than RMI and MBeans for performance
and flexibility reasons
• JDBC optimized for transfer of result set data
• JDBC already abstracts much of database
implementation details
• Eliminate the RMI step, increase performance, and
reduce complexity for Database developers
• Use database security for DM rather than custom
J2EE security
slide 26
Driver Specifics
 Performance
• Result set batching
 Data Types
 Schema Information Retrieval
 Performance information retrieval
slide 27
Key Modifications: Caching
Changes
 Instead of purging results as soon as they are retrieved,
associates results with query history and keep around
as long as possible
slide 28
Key Modifications: Session
Management
 Connection and data persistence across web service
calls
 Get UUID back on login to identify session
• UUID generator security (Java randomUUID for
cryptographically strong randomness)
 Web Services don’t usually save state
• In this case, UUID tied to JHU Query Manager
session
slide 29
Test Plan
 Initial test plan draft developed
 Includes more test details than existing DRL test plan
 Key realization:
• Need to define minimum requirements for
integration of new Data Managers
• Data Manager acceptance testing needed
 Will be updated as we continue to make design
decisions and as software is implemented
slide 30
Performance Testing
 Performance Critical Components
• Result Set Persistence
– Stream large result sets directly to/from disk
• HTTP for Data Transfer
– Zip Data Compression
• JDBC Drivers
– Optimize use of JDBC Driver
• Server Threading
– Test with many distributed clients downloading
– Connection persistence across web service calls
• Division of machines / processes
slide 31
Software Delivery
 Implementation will be provided using a version of the
SDSS database as an example backend. MySQL and
PostgreSQL will also be supported.
 Example WBI will be provided with software
 Example Java and .NET client applications will also be
provided
 Automated test suite will also be delivered
slide 32
Schedule
slide 33
Status
 Completed review of specification
 No DRL problems identified
 Technologies chosen for implementation
• Tomcat
• Axis2 Web Service
• Java Caching System for Result Set caching
• Microsoft JDBC SQL Server Driver
 Initial web service proof of concept developed
 Draft test plan document
slide 34
Demo
slide 35
The PSPS Web Based Interface & User
Interface Clients
Jim Heasley
slide 36
The WBI
 The WBI provides an interface for a human user to request
published science data from the data stores via the DRL. It is one
example of a Published Data Client (PDC). Note that there can be
more than one PDC providing the same functionality.
 The WBI provides both a Standard User API and an Administrative
User API.
 The WBI is in fact a combination of the infrastructure needed to
connect to the DRL and some number of clients that access the
PSPS data stores via the DRL.
 Driving requirement – SECURITY
• Preventing unauthorized access
• Not about encrypting data for privacy!
 REALITY CHECK – it’s a web server with some clients attached!
slide 37
WBI Components
slide 38
WBI Components
 The WBI Software Design Description is in the SAIC generated
document PSDC-630-WBI-SDD
 The WBI Components are
• Administrative User View
– The user interface for an authenticated WBI administrator
• Administrative Web Service Driver
– Programming interface that converts a method call to its
mapped SOAP message and sends it to the DRL.
– There is a 1-to-1 mapping of requests handled by the
Administrative Service Driver to SOAP messages defined in
the DRL WSDL
– Documented in PSDC-630-DRL-PDC-Private-ICD.html
• Request Controller
– Provides stateful management of user requests that may
persist longer than a user’s WBI session.
slide 39
WBI Components
 The WBI Components (continued):
• Standard User View
– Provides the user interface to an authenticated nonadministrative WBI user.
• Standard User Web Service Driver
– Provides a programming interface that converts a method
call to its mapped SOAP message and transmits it to the
DRL. It also performs reverse function for responses/faults
received from the DRL. There is a 1-to-1 mapping of
request to the SOAP messages defined in the DRL WSDL.
– Documented in PSDC-630-DLR-PDC-Public-ICD.html
• WBI Account Manager
– Responsible for authenticating users and granting access
permissions to WBI functionality. Users will be identified by
a user name and password which serves as the
authentication credential.
slide 40
WBI Components
 The WBI Components (continued):
• WBI Log Configuration Manager
– Permits an administrator to define logs, define the
level & verbosity of event reporting, and identify
events reported to administrators.
• WBI Log Manger
– Initializes logs on startup as defined in a
configuration file.
– Coordinates logs from multiple WBI components
to ensure only level of logging specified is done
slide 41
WBI Detailed Design
 Main challenge – negotiation of the Web Services Interface to the
DRL.
 These web services make use of concepts outside the realm
familiar to traditional scientific programers, e.g.,
• XML
• SOAP
• WSDL
• X.509 Certificates
• Digital signatures
 To simplify access for the WBI and other PDCs the Standard Web
Service Driver has been encapsulated in an optional Java-based
component named PDC-Core.
 Documentation of this reference implementation is provided in the
SAIC generated document
PSDC-630-WBI-SDD-Addendum-A_Detailed-Design.html
slide 42
WBI User Interfaces
 The components described to this point are there to
provide the low-level functionality necessary to ask for
and return data from the DRL and the data managers
which connected to it. The astronomers won’t interact
with them directly.
 The USER INTERFACES are the web applications that
use the web services provided by the WBI and DRL are
the tools with which the astronomers will interact.
 As mentioned yesterday, we have followed the advice of
the PDR committee and are providing access via
“recycled” web applications (the SDSS Casjobs web
interface, hereafter Query Manager = Qm), reused tools
(from the MOPS), and a work-alike clone of another
existing web app (IPAC’s Gator).
slide 43
The SDSS Casjobs Web Interface
slide 44
PS1 Casjobs Interface = Qm
slide 45
A PS1 Menu Driven Web
Application
 Following the PDR, I developed a prototype of a menu
driven web application for accessing the tables in the
PS1 data base, modeled on the IPAC’s Infrared Science
Archive Gator interface.
 This application was developed using PHP, a serverside HTML embedded scripting language. There are
PHP APIs available for most major databases.
 The user interface allows generating SQL commands
from scratch in a roll-your-own window or automated
SQL generation from check box selection of database
attributes and user specified spatial constraints.
 The interface is configured using information stored in a
MySQL database. This allows easy modification of
schema, help files, etc.
slide 46
slide 47
Menu Driven Web Interface Modules
Menu Driven Queries
Main Window
schema
sqlpage
catalogs
collections
makeMenu
glossary
Roll your own SQL queries
generateSQL
submitSQL
slide 48
Crocodile Demo
 This first demonstration shows Crocodile set up to use
the 2MASS point source and extended source
databases, the USNO UCAC astrometric catalog, the
UCAC bright star supplemental catalog, the USNO-B
catalog, and the Tycho 2 catalog. (Only 1% of the
2MASS PSC is implemented for the demo, along with
the 2MASS XSC and the UCAC bright star supplement.)
 This second demonstration shows an implementation of
the Crocodile user interface configured to use an early
version of the Pan-STARRS schema. There’s no back
end database attached to this demo.
slide 50
MOPS Tools
 Within the MOPS subsystem the group (in particular Larry
Denneau) has developed an extensive set of software tools for
interacting with the MOPS DB. As the SSDM will be a copy (and hot
spare) of the MOPS DB, these tools can be interfaced to the WBI to
provide access to the SSDM for use by astronomers without
impacting the MOPS functional DB.
 As Larry noted yesterday, the MOPS tools have been developed in
PERL.
 The next 3 slides are screen shots of
• A summary page of a single night’s coverage with links to
additional information
• A section of the tracklet and linking efficiency page for a
lunation
• An a (semimajor axis) vs. e (eccentricity) plot of orbits
discovered by MOPS
slide 51
MOPS Tools
slide 52
MOPS Tools
slide 53
MOPS Tools
slide 54
WBI Hardware Plan
 The anticipated hardware configuration for the WBI component of
the PSPS includes:
• 2 servers to provide a full Windows common infrastructure
• 2 web servers to host the WBI components and possibly
copies of the DRL software
• A network switch (over the entire PSPS) that provides load
balancing for the web servers
• Mass storage (~several TBytes) to store output from the
various DMs.
 Software implementation will include
• Common infrastructure for Windows configuration
• Windows 2003 servers
• IIS web server software
• Windows automated patch installation software
slide 55
WBI Risk Assessment
 Overall Risk Assessment for this subsystem is
LOW!
Because
– The APIs are very well defined and use well
known (in the computer community) software
definitions.
– We are recycling and reusing existing software
and creating a clone of a successful scientific
database interface
– It’s only a damn web server system—it’s NOT
rocket science.
slide 56
PSPS External Interfaces
Jim Heasley
slide 57
PSPS External Interfaces
 The PS1 PSPS will (initially) receive data from the two
data managers being developed under the PS1
construction program: the ODM & MOPS. The external
interfaces defined for these two subsystems will provide
the basic template for connecting future databases and
data stores (e.g., image archives) to the PSPS.
 Actual data transfer will be done via the Data Store
mechanism described yesterday morning.
 For each data store interfaced to a DM we define:
• An Interface Requirements Specification (IRS) that
tells what we must transfer.
• An Interface Control Document (ICD) that tells how
the transfer takes place and the details of what is
actually transferred.
slide 58
PSPS External Interfaces
 We have documented these interfaces as follows:
• IPP-PSPS
– PSDC 930-006: The Interface Requirements Specification
– PSDC 940-006: The Interface Control Document
• MOPS-PSPS
– PSDC 930-007: The Interface Requirements Specification
– PSDC 940-007: The Interface Control Document
 For brevity, in this section I will only discuss the interfacing between
the PSPS and the IPP because
• The MOPS-PSPS interfacing is much simpler in that we are
moving a dump of the MySQL database held within the MOPS.
• The IPP-PSPS interface is more complicated than that for
MOP-PSPS
slide 59
The PS1 IPP-PSPS IRS
 Defined IPP Data Products
• Predefined Objects determined by the IPP during
commissioning and testing.
• Sky Cell Definitions
• Camera Configuration Information
• Detection Attributes common to P2, P4Σ, P4Δ, and image
stack sources
• Detection Attributes common to P2, P4Σ , P4Δ sources when
PSF fitting is insufficient
• Attributes for non-psf sources in image stacks for sources that
are not well described by a PSF fit.
• P2, P4Σ, P4Δ photometry transformation metadata
• Frame Descriptor Metadata (per exposure)
• P2 Image Metadata (per OTA per exposure)
slide 60
The PS1 IPP-PSPS IRS
 Defined IPP Data Products (continued)
• P4Σ, P4Δ, Image Stack Metadata
• Definitions of alternative source fitting models used
on the P2, P4Σ, P4Δ sources that don’t fit the PSF
• List of Survey Programs
• List of Photo-Z recipes
 Defined PSPS (derived) Data Products
• Object Attributes/Statistics
 IPP Data Publication Process
 PSPS Communication Process
slide 61
The PS1 IPP-PSPS ICD





slide 62
The IPP-PSPS ICD describes the details of the publication process of data
moving from IPP to PSPS.
The IPP defines a 1 time transfer to catalog objects from IPP to PSPS to
initialize the objects in the ODM. This is done to jump start the ODM and
detection-to-object correlation process.
The PSPS publication cycle consists of a job, which itself consists of a set
of batches of FITS files and a batch manifest file
The monthly publication process has 4 stages:
• Verification that the FITS files match the batch manifest. If not send,
Batch Manifest Error Notification.
• Verification of the integrity of FITS files via checksums. If not, send
Batch Corruption Error Notification.
• Verification that the content of the FITS files match the manifest
description. If not, send Batch Content Error Notification.
• Detailed verification of every datum in each FITS file. If not, send a
Batch Verification Error Notification.
When IPP has shipped all the batches for a job it sends PSPS a batch
manifest file.
The PS1 IPP-PSPS ICD
 If any error notification is sent the appropriate correction process for
IPP is defined by the ICD.
 The ICD also describes the situations that could give rise to a
broken publication cycle and the actions to be taken to
resynchronize the publication process should that occur.
 Note: times for publication cycle are based on an old (SAIC)
conservation ingest rate. We now expect to do this much faster and
may be able to publish on a faster time scale.
 Data transfer from IPP to PSPS is assumed to occur in a trusted
environment. No authentication measures are specified.
 Data moves as a push from IPP to PSPS. The interface between
the systems is supplemented by event notifications via email from
PSPS to the IPP.
slide 63
The PS1 IPP-PSPS ICD
 The following notifications from PSPS to IPP are
defined:
•
•
•
•
•
•
•
•
•
•
•
slide 64
Batch Manifest Error
Batch Corruption Error
Batch Received
Batch Refused Error
Batch Content Error
Batch Verification Error
Job Manifest Error
Job Verification Error
Job Verification Successful
Job Ingest Failure
Job Publication
(can’t accept data)
(manifest lists batches not found)
(data from different batches inconsistent)
(data verified but PSPS can’t ingest it)
The PS1 IPP-PSPS ICD
 Manifest files are simple XML files describing the
appropriate structure (e.g., batch or job).
 The ICD describes the structure and format of the FITS
files used to transmit the IPP data products.
 The ICD describes in detail the FITS “cards” used in the
Header Data Units (HDU) for each component of the
FITS files used to transfer the IPP data products.
 The data attributes themselves are included in the FITS
files as binary FITS tables as defined in the appropriate
HDU.
slide 65
The PSPS Test Plan
Jim Heasley
slide 66
Test Plan Overview
 SAIC designed a PSPS Software Test Plan (STP) that is presented
in PSDC-630-STP. While the document was prepared at a time
when SAIC was the primary support for the PSPS development
(and hence is worded to reflect their expectation to be doing the
code development around an Oracle database system and the
subsequent testing), the test plan itself is very generic and
database independent.
 As with the previous discussion of the PSPS external interfaces, we
will concentrate here on the testing of the ODM.
 The primary subsections of the STP cover
• Test environment
• Test identification
• Test Schedule
slide 67
PSPS Software Test Environment
slide 68
Software Test Environment
 The software test environment is designed to support:
• Design validation
• Component testing
• Integration testing
• Deployment testing
 PSPS Test Software will include the following tools:
• DM simulator
• DRL simulator
• IPP simulator
• MOPS simulator
• WBI simulator
• WBI user simulator
slide 69
Test Data from IPP
 The IPP will need to provide test data sets for all the
data coming to the PSPS as defined in the ICD. In
particular, for the ODM we will need samples of:
• Alternate Source Model Fits
• Camera configuration
• Image stack (cumulative sky) data
• P2 Frame data
• P4Δ high significance data
• P4Δ low significance data
• Photometry transformation information
• Predefined objects from IPP
• Sky cell definitions
• Survey program definitions
slide 70
Test Data Sets
 The STP defines test data sets that are jobs and
batches of the IfA provided data sets that are used to
evaluate the ingest mode of the ODM. In these tests one
or more of the sample data sets are corrupted to test
whether the software can identify the data values that
are inappropriate or out of range.
slide 71
Test Identification
 Test Levels
• Design validation
– Tests to be peformed as part of the development process
during the design an code/unit test phase.
• Component testing
– Tests performed to verify each component meets
specifications
• Integration testing
– Tests conducted by JHU, Referentia, and IfA to confirm the
PSPS components work with each other
• Deployment testing
– Tests to verify PSPS works with IPP and MOPS
slide 72
Planned Tests
 Test cases describe prerequisite conditions, inputs,
expected results, conditions for success, assumptions,
constraints, and outputs to a test log when a test
procedure is performed. One or more test cases are
defined in the Software Test Description for every test
(PSDC-630-STD).
 The test cases are organized by component/subsystem
as follows:
• WBI
• DRL
• DM
• PSPS
slide 73
Real World Testing
 Fact of life: the development timelines for the PSPS
components differ, in particular, the more complex ODM
will be finished later than the others.
 The good news: we already have a working MOPS DB
and Perl based queries, the DRL is making excellent
progress.
 So, I plan to have the first “full” integration testing of the
PSPS done earlier than ODM’s completion by have an
end-to-end testing using the WBI MOPS client, the DRL,
and the SSDM (MOPS clone).
 JHU can provide a MyBestPS1 minisystem for testing
purposes at Referentia, and network access to the ODM
prototype for testing and integration purposes.
slide 74
The PSPS Implementation Schedule
Jim Heasley
slide 75
Schedule
 The three components of the PSPS have rather different
time scales for completion.
 The driving factor for the PSPS implementation
schedule is the ODM work at JHU. You’ve already heard
from that team that we anticipate a complete system
design to be available and tested by September 2008.
 The DRL development work by Referentia is due to be
completed by April 1, 2008.
 No detailed schedule is available on the WBI
infrastructure software (as Mr. Holmberg hasn’t started
work yet). As shown in the previous sections, we
already have working prototypes of all three web access
clients we intend to implement.
slide 76
Schedule – Target Mile Stones








slide 77
Jan. 2008
Jan. 2008
Apr. 2008
Aug. 2008
Sep. 2008
Oct. 2008
Nov. 2008
Dec. 2008
-
Define hardware for Yr 1 configuration
New software engineers start @ IfA, JHU
Referentia finishes DRL programming
JHU Finishes ODM design
Hardware ordered for PSPS
Install hardware at site
System integration
PSPS ORR
Risk Evaluation & Mitigation Plan
Jim Heasley
slide 78
Risk
Paraphrasing an American “philosopher”:
There are the things you know you know,
there are the things that you know you don’t know,
And there are the things that you don’t know that you
don’t know.
D. Rumsfeld
slide 79
Risk
 In this section we want to show that
• We’ve maximized those things in the first category.
• Figured out what we need to do about things in the
second category
• And hopefully minimized those in the third category!
slide 80
Inter-Subsystem Risks
 Telescope, Camera, and Observatory and
Telescope Instrumentation Software Subsystems
•
•
•
slide 81
No direct risk to PSPS schedule or cost
Any delay on their side gives schedule relief to PSPS
integration
Overall risk to PSPS - LOW
Inter-Subsystem Risks
 Moving Object Processing Software (MOPS)
•
•
•
•
slide 82
The Solar System Data Manager (SSDM) in PSPS will
import much simpler data products than those provided
by the IPP.
The SSDM is will be clone of the MOPS internal DM.
MOPS decides what it exports.
Maturity of MOPS-PSPS interface does not pose
significant risk to PSPS development.
Overall risks to PSPS - LOW
Inter-Subsystem Risks
 Image Processing Pipeline (IPP)
•
The reality requirement: PSPS must ingest the data products
from IPP at the nominal monthly processing rate… whatever
it turns out to be
• IPP is the stressing data source for the PS1 PSPS
• IPP is ahead of PSPS in the design/review process
– Some fine tuning of the data attributes and file formats
may be needed to reach a final configuration
• The good news is that IPP will be processing data for ~6
months before it is “ready” for delivery to the PSPS, so we
will have ample opportunity to test both simulated and real
data.
 Risk Level – LOW, provided we are able to iterate/refine issues
with the IPP.
slide 83
Intra-Subsystem Risks




slide 84
WBI – LOW
DRL – LOW
SSDM – LOW
ODM –LOW to MODERATE
General Areas of Risk


slide 85
Science Rules
• Over the course of these presentations there have been some points
that have been left unspecified on purpose, e.g. radius size for
detection-object correlation, number of detections to create an object,
etc. These will ultimately be determined by experimentation with real
data from the IPP.
• We believe this is a LOW risk issue provided IPP can provide to us
the appropriate data before system deployment so that we can
conduction the appropriate experiments and deduce the rules.
Short time scale versions
• This is the getting the same answer as you did from an old query in a
database which is constantly being updated (and hopefully improved)!
• As noted the previously we hope to deal with this by providing snap
shots of the object table and by providing adequate information in the
schema so that one could recover to the state the database was in
when the query was made.
• Issues generating risk are how often are the snap shots made, how
long do they persist, how easy and/or fast does the process of
rerunning an old query have to be? Risks here involve cost of storage
for the snap shots and perhaps programming effort.
• Risk is LOW unless “unreasonable” demands are placed on the ODM.
General Areas of Risk
 Major data recalibrations and versions
• At the end of the 1st year of operation the PS1 team intends to
develop new astrometric and photometric calibrations to the
data derived from the data in hand at that time.
• Such new calibrations can potentially change every derived
number in the database for both detections and objects.
• Experience with SDSS indicates that astronomers will also
want to keep the “DR1” of the database around as they will
have written papers based upon that calibration. Full
versioning of the old tables poses a COST RISK.
• It may be feasible to solve this problem by providing correction
tables for new calibrations and presenting the different
calibrations to users as database views. While this should work
in principle, it has not been tested in practice and hence
represents a MODERATE risk.
slide 86
General Areas of Risk
 Dealing with crawlers and value added data generators
• The IPP and PSPS have now agreed that the value added
products will be calculated by teams from the science
consortium and not be performed within the database.
• We have three potential schemes available to serve these
products back to users
– By providing tables in the ODM into which user provided
data could be loaded if it is deemed to be especially
desirable to serve it this way. These data would be loaded
on a schedule that coincides with regular ODM updates.
– As client web interfaces and DMs attached to the DRL.
– PS1 Consortium clients could have their own MyDB space
associated with the ODM (either within the project storage
area or as add on storage paid for by the client) and their
tables could be exported to other PSPS ODM users via the
Qm group mechanism.
• Overall risk to PSPS is LOW (after initial efforts to get the
process going).
slide 87
General Areas of Risk
 Geoplexing
• We need to be able to distribute incremental updates of the
ODM to other sites off Maui in order to
– Maintain a physically distinct (and hence) safe backup of
the ODM in case of catastrophic events at the primary ODM
site.
– Have a mechanism to provide data to PS1 Consortium
partners for their ODM mirrors.
• This process is ultimately limited by the bandwidth available
between Maui and the alternate sites.
• Risk level is LOW to MODERATE for providing regular
incremental updates but HIGH if one needed to recreate the
primary site from one of the remote ones.
slide 88
General Areas of Risk
 Location, Location, Location!
• The site for the PSPS has not been determined, with
both the IfA buildings on Maui and Oahu being
possible choices.
• Strictly speaking this is an operational issue (with
some issues that involve the PS1 Consortium) that
doesn’t need to be addressed at CDR, but this
uncertainty does have implications on the design!
• Risk assessment– MODERATE to HIGH?
slide 89
General Areas of Risk
 Staffing
• Software development staffing
– We have adequate personnel to deal with the WBI and DRL
development. The ODM situation is tighter but will be
helped with the addition of two FTE provided by the PS
Project in January.
– Risk – LOW to MODERATE
• Operational Staffing
– To date there has been no clear discussion of the
requirements for an operational staff to install or administer
the PSPS system. This was largely because until the
completion of the CDR we do not know exactly what we’re
planning for.
– The JHU team advises that we need to have 2 people
trained to perform the hardware installation and system
admin functions.
– Risk –MODERATE as it’s budget issue, but unclear to me
whether it belongs to the Project or the Consortium (or
both).
slide 90
General Areas of Risk
 PSPS stability/trouble shooting/testing/upgrades
• Good engineering practice dictates that one does
not do testing for stability, trouble shoot, or upgrades
on a production system.
• We believe that to be able to address the problems
that are almost certainly going to arise with a new
system like the ODM we will require a testing
system for this purpose.
• This system was not anticipated in the original
PSPS plan but should fit within the maximum
possible budget available to the subsystem. Without
this test system overall risk to PSPS would be
HIGH.
slide 91
General Areas of Risk
 PSPS stability/trouble shooting/testing/upgrades
• Good engineering practice dictates that one does
not do testing for stability, trouble shoot, or upgrades
on a production system.
• We believe that to be able to address the problems
that are almost certainly going to arise with a new
system like the ODM we will require a testing
system for this purpose.
• This system was not anticipated in the original
PSPS plan but should fit within the maximum
possible budget available to the subsystem. Without
this test system overall risk to PSPS would be
HIGH.
slide 92
General Areas of Risk
 Budget
• My concern here is not what it was at PDR where it
was unclear what the project was willing to commit
to the PSPS development, but rather a realistic
worry that if some other critical subsystem suffers a
catastrophic problem funds targeted to PSPS will be
diverted for that purpose. At that point we will have
to make some cutbacks in what we want to do with
the PSPS system.
• Risk level ???
slide 93
In Conclusion
slide 94
Does Our Design Meet the Top
Level Design Requirements?
 3.3.01 The PSPS shall be able to ingest a total of 1.5x1011 P2
detections, 8.3x1010 cumulative sky detections, and 5.5 x109
celestial objects together with their linkages.
• Yes. Prototyping efforts for the ODM show this target can be
achieved.
 3.3.02 The PSPS shall be able to ingest the observational
metadata for up to a total of 1,050,000 observations.
• Yes. Prototyping efforts for the ODM show this target can be
achieved.
 3.3.0.3 The PS1 PSPS shall be capable of archiving up to ~ 100
Terabytes of data (TBR).
• Yes. Prototyping efforts for the ODM show this target can be
achieved.
 3.3.0.4 The PSPS shall archive the PS1 data products.
• Yes. The design provides sufficient redundancy that the
products can not only be input into the database but also
reliably maintained in case of hardware failures.
slide 95
Does Our Design Meet the Top
Level Design Requirements?




slide 96
3.3.0.5 The PSPS shall possess a computer security system to protect
potentially vulnerable subsystems from malicious external actions.
• Yes. Multiple layers of security are built in via the WBI, the DRL, and
the Data Stores.
3.3.0.6 The PSPS shall provide end-users access to detections of objects
in the Pan-STARRS databases.
• Yes. Multiple WBI interfaces are available to provide interactive user
access as well as allowing for machine based access by Published
Data Clients.
3.3.0.7 The PSPS shall provide end-users access to the cumulative
stationary sky images generated by the Pan-STARRS.
• Yes. For PS1 the PSPS will serve as a front end for requests for
images from the IPP.
3.3.0.8 The PSPS shall provide end-users with metadata required to
interpret the observational legacy and processing history of the PanSTARRS data products.
• Yes. The relevant metadata are transferred fro the IPP along with the
data products and the schema of the ODM provides for their access.
Does Our Design Meet the Top
Level Design Requirements?




slide 97
3.3.0.9 The PSPS shall provide end-users with Pan-STARRS detections of
objects in the Solar System for which attributes can be assigned.
• Yes. These are contained in the MOPS DB and the SSDM component
of the PSPS is a “hot spare” of the former.
3.3.0.10 The PSPS shall provide end-users with derived Solar System
objects deduced from Pan-STARRS attributed observations and
observations from other sources.
• Yes. These are contained in the MOPS DB and the SSDM component
of the PSPS is a “hot spare” of the former.
3.3.0.11 The PSPS shall provide the capability for end-users to construct
queries to search the Pan-STARRS data products over space and time to
examine magnitudes, colors, and proper motions.
• Yes. This will be possible through multiple mechanisms attached to
the DRL through several web clients.
3.3.0.12 The PSPS shall provide a mass storage system with a reliability
requirement of 99.9% (TBR).
• Yes. The ODM design provides sufficient redundancy (based on
experience with the SDSS database) to reach this level of reliability.
Does Our Design Meet the Top
Level Design Requirements?
 3.3.0.13 The PSPS baseline configuration should
accommodate future additions of databases (i.e., be
expandable).
• Yes. The basic design of the DRL provides for the
possibility to add both new published data clients
and data stores. Further, we have multiple
approaches to incorporate “value added” data
products within the PSPS design.
slide 98
Does the PSPS Design Meet the
CDR Criteria?
 Has this design we’ve presented matured to the CDR
level?
• We believe this is the case.
• Indeed all other PS1 subsystems that passed CDR
have experienced continuing development. That’s
not an excuse, just a statement of fact.
 Is the design reliable and maintainable?
• In our opinion, yes.
 Is the hardware within the scope of the project budget?
• The current design uses generally available
commodity hardware. We continue to seek
leveraged hardware acquisition with interested
vendors.
slide 99
Does the PSPS Design Meet the
CDR Criteria?
 Is the test plan well conceived?
• Yes.
 Are the interface designs mature?
• The interfaces to other PS1 subsystems are in fact stable and
have changed little since PDR. That doesn’t mean that some
“tweaking” won’t be needed.
• The Data Store component is already used by other PS1
subsystems.
 Have the risks been identified and mitigation strategies spelled out?
• Yes, we believe so.
• Also note that of the risks we have identified none are fatal to
the basic design or cause us to not meet any requirements.
These risks are in many ways related to convenience to the
PSPS users.
slide 101
Finally
 Hopefully the very bright folks I’ve been working with at
JHU and Referentia (and previously at SAIC) have
found solutions for the hard problems and anticipated
the ones which might become hard.
 Now, one of your tasks as the CDR panel is to help us
by telling us what we don’t know that we don’t know.
 And so, for patiently listening to us for the past 2 days
slide 102
Mahalo!
slide 103