GLAST LAT Project ISOC Peer Review - March 2, 2004 Document: LAT-PR-03213-01 Section 4.3 Pipeline, Data Storage and Networking Issues1 Gamma-ray Large.

GLAST LAT Project ISOC Peer Review - March 2, 2004

Document: LAT-PR-03213-01 Section 4.3 Pipeline, Data Storage and Networking Issues 1

Gamma-ray Large Gamma-ray Large Area Space Area Space TelescopeTelescope

GLAST Large Area TelescopeGLAST Large Area Telescope

ISOC Peer Review Section 4.3

Pipeline, Data Storage and Networking Issuesfor the ISOC

Richard DuboisSAS System Manager



OutlineOutline

• SAS Summary Requirements

• Pipeline

• Requirements

• Processing database

• Prototype status

• Networking

• Proposed Network Topology

• Network Monitoring

• File exchange

• Security

• Data Storage and Archive



Level III Requirements SummaryLevel III Requirements SummaryRef: LAT-SS-00020

Function Requirement Expected Performance (if applicable)

Verification

perform prompt processing from Level 0 through Level 1

keep pace with up to 10 GB Level 0 per day and deliver to SSC within 24 hrs

demonstration

provide near-real time monitoring to IOC

within 6 hrs demonstration

maintain state and performance tracking

demonstration

facilitate monitoring and updating of iinstrument calibrations

demonstration

archive all data passing through

> 50 TB on disk and tape backup

demonstration

Flight Ground Processing

• Basic requirements are to routinely handle ~10 GB/day in multiple passes coming from the MOC

• @ 150 Mb/s – 2 GB should take < 2 mins

• Outgoing volume to SSC is << 1 GB/day

• NASA 2810 Security Regs – normal security levels for IOC’s; as practiced by computing centers already



Pipeline SpecPipeline Spec

• Function– The Pipeline facility has five major functions

• automatically process Level 0 data through reconstruction (Level 1) • provide near real-time feedback to IOC • facilitate the verification and generation of new calibration constants • produce bulk Monte Carlo simulations • backup all data that passes through

• Must be able to perform these functions in parallel• Fully configurable, parallel task chains allow great flexibility for use online as

well as offline– Will test the online capabilities during Flight Integration

• The pipeline database and server, and diagnostics database have been specified (will need revision after prototype experience!)

– database: LAT-TD-00553 – server: LAT-TD-00773 – diagnostics: LAT-TD-00876



Expected CapacityExpected Capacity

• We routinely made use of 100-300 processors on the SLAC farm for repeated Monte Carlo simulations, lasting weeks– Expanding farm net to France and Italy– Unknown yet what our MC needs will be– We are very small compared to our SLAC neighbour BABAR –

computing center sized for them• 2000-3000 CPUS; 300 TB of disk; 6 robotic silos holding

~30000 200 GB tapes total

– SLAC computing center has guaranteed our needs for CPU and disk, including maintenance for the life of the mission.

– Data rate small compared to already demonstrated MC capability• ~75 of today’s CPUs to handle 5 hrs of data in 1 hour @ 0.15

sec/event• Onboard compression may make it 75 of tomorrow’s CPUs too



Pipeline in PicturesPipeline in Pictures

State machine + complete processing record

Configurable linked list of applications to run

Expandable and configurable set of processing nodes



Processing Dataset CatalogueProcessing Dataset Catalogue

Datasets grouped by task

Datasets grouped by task

Datasets info is hereDatasets info is here

Processing records

Processing records



First Prototype - OPUSFirst Prototype - OPUS

Open source project from STScI

In use by several missionsNow outfitted to run DC1 dataset

OPUS Java mangers for pipelines



Disk and ArchivesDisk and Archives

• We expect ~10 GB raw data per day and assume comparable volume of events for MC– Leads to ~50 TB over 5 years for all data types

• No longer frightening – keep it all on disk

– Use SLAC’s mstore archiving system to keep a copy in the silo

• Already practicing with it and will hook it up to OPUS

– Archive all data we touch; track in dataset catalogue

– Not an issue



Network Path: SLAC-GoddardNetwork Path: SLAC-Goddard

٭

٭

٭

٭

٭

SLAC Stanford Oakland (CENIC) LA UC-AID (Abilene) Houston Atlanta Washington GSFC (77 ms ping)



ISOC Stanford/SLAC NetworkISOC Stanford/SLAC Network

• SLAC Computing Center – OC48 connection to outside world– provides data connections to MOC and SSC – hosts the data and processing pipeline– Transfers MUCH larger datasets around the world for

BABAR– World renowned for network monitoring expertise

• Will leverage this to understand our open internet model– Sadly, a great deal of expertise with enterprise security as

well• Part of ISOC expected to be in new Kavli Institute building on

campus– Connected by fiber (~2 ms ping)– Mostly monitoring and communicating with processes/data

at SLAC



Network MonitoringNetwork Monitoring

Need to understand failover reliability, capacity and latency



LAT MonitoringLAT Monitoring

Monitoring nodes at most LAT collaborating institutions

LAT Monitoring

Keep track of connections to collaboration sites

Alerts if they go down

Fodder for complaints if poor connectivity



File ExchangeFile Exchange

• Secure – No passwords in plain text etc

• Reliable – Has to work > 99% of the time (say)

• handle the (small) data volume – order 10 GB/day from Goddard (MOC); 0.3 GB/day back to Goddard (SSC) – keep records of transfers – database records of files sent and received – handshakes – both ends agree on what happened – some kind of clean error recovery – Notification sent out on failures

• Web interface to track performance

• Are investigating DTS now– Not super happy with how it’s built as far as reliability and security go– Installing it now– Starting to work with author on issues

• To be captured in an ICD with the SSC prior to DC2



SecuritySecurity

• Network security – application vs network– ssh/vpn among all sites – MOC, SSC and internal ISOC– A possible avenue is to make all applications secure (ie

encrypted), using SSL.

• File and Database security– Controlled membership in disk ACLs– Controlled access to databases– Depend on SLAC security otherwise



SummarySummary

• We are testing out the OPUS pipeline as our first prototype– Already outfitted to run DC1 dataset– Interfaces to processing database and SLAC batch done– Practice with DCs and Flight Integration

• We expect to need O(50 TB) of disk and ~2-3x that in tape archive– Not an issue

• We expect to use Internet2 connectivity for reliable and fast transfer of data between SLAC and Goddard– Transfer rates of > 150 Mb/s already demonstrated– < 2 min transfer for standard downlink. More than adequate.– Starting a program of routine network monitoring to practice

• Network security is an ongoing, but largely solved, problem– There are well-known mechanisms to protect sites– We will leverage considerable expertise from the SLAC and

Stanford networking/security folks

GLAST LAT Project ISOC Peer Review - March 2, 2004 Document: LAT-PR-03213-01 Section 4.3 Pipeline, Data Storage and Networking Issues1 Gamma-ray Large.

Documents

slac security

internal isoc

opus pipeline

fast transfer of data

networking issues16

goddard ssc

small data volume order

archive slide