INFSO-RI-508833 Enabling Grids for E-sciencE The gLite File Transfer Service: Middleware Lessons Learned form Service Challenges Paolo.

INFSO-RI-508833

Enabling Grids for E-sciencE

www.eu-egee.org

The gLite File Transfer Service:Middleware Lessons Learned formService Challenges

Paolo BadinoOn behalf of the JRA1 Data Management team

CHEP06 - Mumbai, 14 February 2006

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 2

Enabling Grids for E-sciencE

INFSO-RI-508833

Outline

• The gLite File Transfer Service – in LCG Service Challenge 3– Architecture– Administration– Security

• Result

• LCG Service Challenge 4

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 3

Enabling Grids for E-sciencE

INFSO-RI-508833

gLite File Transfer Service

• Low Level, point to point data movement service• Part of the gLite Middleware• Asynchronous transfers• Based on Globus GridFTP, SRM• Retry policies

• Not just a Reliable Transfer Service

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 4

Enabling Grids for E-sciencE

INFSO-RI-508833

LCG Service Challenges

• Production Quality• Focus on Service:

– Network service– Robust file transfer service– Sustainability

• Heterogeneous Environment– Castor2, dCache, DPM

• Service Challenge 2– Radiant

• Service Challenge 3– FTS (gLite 1.4.1)– CMS Phedex

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 5

Enabling Grids for E-sciencE

INFSO-RI-508833

The Key Concept: The Channel

• Logical unit of management– Represent a directed network pipe between two sites

• Mono-directional• Independently manageable

– State– Number of streams – Number of concurrent transfers

• Inter-VO scheduling– VO share

• No Routing

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 6

Enabling Grids for E-sciencE

INFSO-RI-508833

FTS Architecture

• FTS Web Service– User, Administration and Monitoring Interfaces

• File Transfer Queue

• File Transfer Agents– VO Agents– Channels Agents

• SOAP API and CLI

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 7

Enabling Grids for E-sciencE

INFSO-RI-508833

Transfer Jobs and Files

• Job– Represent the transfer request– Identified by a GUID

• File– source-destination file names pair

Job States File States

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 8

Enabling Grids for E-sciencE

INFSO-RI-508833

Channel Administration

• Channel has a number of properties– State

Active Inactive Drain Stopped Halted

– Number of concurrent files transfers– Number of streams– VO Share

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 9

Enabling Grids for E-sciencE

INFSO-RI-508833

Channel Agent Properties

• Transfer Type– SRM Get & Put + 3rd party GridFTP – SRM Copy

• Timeouts– SRM interactions– Transfer– Transfer Markers

• TCP Buffer Size

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 10

Enabling Grids for E-sciencE

INFSO-RI-508833

Security: Roles

• Submitter User• Administrator• Vetoed User

• Channel Manager• VO Manager

• Gridmap files• VOMS roles

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 11

Enabling Grids for E-sciencE

INFSO-RI-508833

Security: Delegation

• Delegated User Credentials retrieved from MyProxy

• Agents are Authorized Retrievers– Enough for Production– … problematic for Analysis

• Delegation PortType

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 12

Enabling Grids for E-sciencE

INFSO-RI-508833

Monitoring

• Provides statistic information based on the FTS behavior and usage– Snapshots of number of requests in a given state– Average rate and bytes written in a given period

• Grouped by VO or Channels

• Channel Monitoring

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 13

Enabling Grids for E-sciencE

INFSO-RI-508833

FTS Web Interface

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 14

Enabling Grids for E-sciencE

INFSO-RI-508833

Retry Logic

• Simple retry strategy– Retry all the files up to 3 times, independently of the error reason

• Complex retry strategy (gLite 1.5)– Evaluate the transfer attempts history – Resilient to failures in the error handling

• Experiments’ specific retry logic (gLite 1.5)

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 15

Enabling Grids for E-sciencE

INFSO-RI-508833

What SC achieved so far

• SC3 Rerun (January 2006)• All sites achieved target rate• 8/11 sites achieved nominal rate

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 16

Enabling Grids for E-sciencE

INFSO-RI-508833

Preparing for SC4

• Stability• Sustainability• Robustness

• Monitoring• Procedures• SRM v2• Integration with experiments’ frameworks

– Hooks/Catalog plug-ins

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 17

Enabling Grids for E-sciencE

INFSO-RI-508833

Summary

• gLite FTS – Reliable Transfer Service– Transfer Management Tool

• LCG Service Challenge 3– Architecture

Channels

– Security– Feedbacks from experiments– Results achieved

• FTS in LCG Service Challenge 4

The gLite File Transfer Service: Middleware Lessons Learned from the Service Challenges 18

Enabling Grids for E-sciencE

INFSO-RI-508833

Links

• gLite– http://www.glite.org/

• gLite JRA1 Data Management– http://egee-jra1-dm.web.cern.ch/egee-jra1-dm/

• Service Challenges Wiki– https://twiki.cern.ch/twiki/bin/view/LCG/LCGServiceChallenges

• FTS Wiki– https://twiki.cern.ch/twiki/bin/view/EGEE/FTS/