Andrea Manzi CERN On behalf of the DPM team HEPiX Fall 2014 Workshop DPM performance tuning hints for HTTP/WebDAV and Xrootd 1 16/10/2014.

Slides:



Advertisements
Similar presentations
Storage Workshop Summary Wahid Bhimji University Of Edinburgh On behalf all of the participants…
Advertisements

Wahid Bhimji SRM; FTS3; xrootd; DPM collaborations; cluster filesystems.
HDFS and S3 plugins Andrea Manzi Martin Hellmich 13/12/2013.
O. Keeble, F. Furano, A. Manzi, A. Ayllon, I. Calvet, M. Hellmich DPM Workshop 2014 DPM Monitoring.
1 Exploring Data Reliability Tradeoffs in Replicated Storage Systems NetSysLab The University of British Columbia Abdullah Gharaibeh Matei Ripeanu.
DPM Italian sites and EPEL testbed in Italy Alessandro De Salvo (INFN, Roma1), Alessandra Doria (INFN, Napoli), Elisabetta Vilucchi (INFN, Laboratori Nazionali.
Client/Server Grid applications to manage complex workflows Filippo Spiga* on behalf of CRAB development team * INFN Milano Bicocca (IT)
LHC Experiment Dashboard Main areas covered by the Experiment Dashboard: Data processing monitoring (job monitoring) Data transfer monitoring Site/service.
A. Mohapatra, HEPiX 2013 Ann Arbor1 UW Madison CMS T2 site report D. Bradley, T. Sarangi, S. Dasu, A. Mohapatra HEP Computing Group Outline  Infrastructure.
StoRM Some basics and a comparison with DPM Wahid Bhimji University of Edinburgh GridPP Storage Workshop 31-Mar-101Wahid Bhimji – StoRM.
INTRODUCTION The GRID Data Center at INFN Pisa hosts a big Tier2 for the CMS experiment, together with local usage from other HEP related/not related activities.
Tier 3 Data Management, Tier 3 Rucio Caches Doug Benjamin Duke University.
Large Scale Test of a storage solution based on an Industry Standard Michael Ernst Brookhaven National Laboratory ADC Retreat Naples, Italy February 2,
Configuration Management with Cobbler and Puppet Kashif Mohammad University of Oxford.
Grid Lab About the need of 3 Tier storage 5/22/121CHEP 2012, The need of 3 Tier storage Dmitri Ozerov Patrick Fuhrmann CHEP 2012, NYC, May 22, 2012 Grid.
Quick Introduction to NorduGrid Oxana Smirnova 4 th Nordic LHC Workshop November 23, 2001, Stockholm.
Predrag Buncic (CERN/PH-SFT) WP9 - Workshop Summary
Optimisation of Grid Enabled Storage at Small Sites Jamie K. Ferguson University of Glasgow – Jamie K. Ferguson – University.
OSG Tier 3 support Marco Mambelli - OSG Tier 3 Dan Fraser - OSG Tier 3 liaison Tanya Levshina - OSG.
Overview of grid activities in France in relation to FKPPL FKPPL Workshop Thursday February 26th, 2009 Dominique Boutigny.
INFSO-RI Enabling Grids for E-sciencE Enabling Grids for E-sciencE Pre-GDB Storage Classes summary of discussions Flavia Donno Pre-GDB.
WebFTS File Transfer Web Interface for FTS3 Andrea Manzi On behalf of the FTS team Workshop on Cloud Services for File Synchronisation and Sharing.
Performance Tests of DPM Sites for CMS AAA Federica Fanzago on behalf of the AAA team.
6/23/2005 R. GARDNER OSG Baseline Services 1 OSG Baseline Services In my talk I’d like to discuss two questions:  What capabilities are we aiming for.
Grid Technology CERN IT Department CH-1211 Geneva 23 Switzerland t DBCF GT DPM Collaboration Motivation and proposal Oliver Keeble CERN On.
SLACFederated Storage Workshop Summary For pre-GDB (Data Access) Meeting 5/13/14 Andrew Hanushevsky SLAC National Accelerator Laboratory.
Grid Technology CERN IT Department CH-1211 Geneva 23 Switzerland t DBCF GT DPM / LFC and FTS news Ricardo Rocha ( on behalf of the IT/GT/DMS.
Grid Technology CERN IT Department CH-1211 Geneva 23 Switzerland t DBCF GT Standard Interfaces to Grid Storage DPM and LFC Update Ricardo.
INFSO-RI Enabling Grids for E-sciencE ARDA Experiment Dashboard Ricardo Rocha (ARDA – CERN) on behalf of the Dashboard Team.
CERN IT Department CH-1211 Geneva 23 Switzerland GT HTTP solutions for data access, transfer, federation Fabrizio Furano (presenter) on.
Grid Technology CERN IT Department CH-1211 Geneva 23 Switzerland t DBCF GT Upcoming Features and Roadmap Ricardo Rocha ( on behalf of the.
Doug Benjamin Duke University. 2 ESD/AOD, D 1 PD, D 2 PD - POOL based D 3 PD - flat ntuple Contents defined by physics group(s) - made in official production.
Data Transfer Service Challenge Infrastructure Ian Bird GDB 12 th January 2005.
EGI-Engage Data Services and Solutions Part 1: Data in the Grid Vincenzo Spinoso EGI.eu/INFN Data Services.
EGI-InSPIRE RI EGI-InSPIRE EGI-InSPIRE RI Monitoring of the LHC Computing Activities Key Results from the Services.
Grid Technology CERN IT Department CH-1211 Geneva 23 Switzerland t DBCF GT Overview of DMLite Ricardo Rocha ( on behalf of the LCGDM team.
DMLite GridFTP frontend Andrey Kiryanov IT/SDC 13/12/2013.
T3g software services Outline of the T3g Components R. Yoshida (ANL)
CernVM-FS Infrastructure for EGI VOs Catalin Condurache - STFC RAL Tier1 EGI Webinar, 5 September 2013.
WLCG and IPv6 David Kelsey (STFC-RAL) LHCOPN/LHCONE, Rome 28 Apr 2014.
Testing Infrastructure Wahid Bhimji Sam Skipsey Intro: what to test Existing testing frameworks A proposal.
EMI is partially funded by the European Commission under Grant Agreement RI Roadmap & Future Work Ricardo Rocha ( on behalf of the DPM team )
INFSO-RI Enabling Grids for E-sciencE File Transfer Software and Service SC3 Gavin McCance – JRA1 Data Management Cluster Service.
Andrea Manzi CERN EGI Conference on Challenges and Solutions for Big Data Processing on cloud 24/09/2014 Storage Management Overview 1 24/09/2014.
Breaking the frontiers of the Grid R. Graciani EGI TF 2012.
XRootD Monitoring Report A.Beche D.Giordano. Outlines  Talk 1: XRootD Monitoring Dashboard  Context  Dataflow and deployment model  Database: storage.
An Analysis of Data Access Methods within WLCG Shaun de Witt, Andrew Lahiff (STFC)
The HEPiX IPv6 Working Group David Kelsey (STFC-RAL) EGI OMB 19 Dec 2013.
WP5 – Infrastructure Operations Test and Production Infrastructures StratusLab kick-off meeting June 2010, Orsay, France GRNET.
Security recommendations DPM Jean-Philippe Baud CERN/IT.
DPM: Future Proof Storage Ricardo Rocha ( on behalf of the DPM team ) EMI INFSO-RI
Grid Technology CERN IT Department CH-1211 Geneva 23 Switzerland t DBCF GT Standard Protocols in DPM Ricardo Rocha.
EMI is partially funded by the European Commission under Grant Agreement RI DPM in EMI-II HTTP and NFS interfaces Oliver Keeble On behalf of DPM.
EMI is partially funded by the European Commission under Grant Agreement RI Future Proof Storage with DPM Oliver Keeble (on behalf of the CERN IT-GT-DMS.
HEPiX IPv6 Working Group David Kelsey (STFC-RAL) GridPP33 Ambleside 22 Aug 2014.
Jean-Philippe Baud, IT-GD, CERN November 2007
WLCG IPv6 deployment strategy
DPM at ATLAS sites and testbeds in Italy
Dynamic Storage Federation based on open protocols
Ricardo Rocha ( on behalf of the DPM team )
Virtualization and Clouds ATLAS position
DPM Installation Configuration
StoRM: a SRM solution for disk based storage systems
Diskpool and cloud storage benchmarks used in IT-DSS
Data Bridge Solving diverse data access in scientific applications
Dynafed, DPM and EGI DPM workshop 2016 Speaker: Fabrizio Furano
StoRM Architecture and Daemons
DPM releases and platforms status
Ákos Frohner EGEE'08 September 2008
DCache things Paul Millar … on behalf of the dCache team.
Presentation transcript:

Andrea Manzi CERN On behalf of the DPM team HEPiX Fall 2014 Workshop DPM performance tuning hints for HTTP/WebDAV and Xrootd 1 16/10/2014

Outline  DPM Overview  Performance Assessment  Internal Testbed  CMS AAA  Performance Tuning  Results  Summary 2 16/10/2014 DPM performance tuning hints for HTTP/WebDAV and Xrootd

Disk Pool Manager (DPM)  The Disk Pool Manager (DPM) is a lightweight storage solution for grid sites. It offers a simple way to create a disk-based grid storage element and supports relevant protocols for file management and access.  It focuses on manageability (ease of installation, configuration, low effort of maintenance), while providing all required functionalities for a grid storage solution:  support for multiple disk server nodes  multiple file replicas in disk pools  multi VOs support  multi protocol: xrootd, HTTP/WebDAV, GridFTP, SRM 3 16/10/2014 DPM performance tuning hints for HTTP/WebDAV and Xrootd

DPM Collaboration  Started in 2013 in order to maintain, develop and support the DPM Software stack for the benefit of the user communities members of the collaboration:  CERN, CESNET, GridPP, Japan, France, Italy and ASGC. 4 16/10/2014 DPM performance tuning hints for HTTP/WebDAV and Xrootd

Deployment statistics  DPM is the most widely deployed grid storage system in EGI and WLCG  Over 200 sites in 50 regions  Over 300 VOs  WLCG statistics:  mon.cern.ch/dashboard/request.py/siteview#currentV iew=Storage+Deployment mon.cern.ch/dashboard/request.py/siteview#currentV iew=Storage+Deployment  ~50 PB (10 sites with > 1PB) 5 16/10/2014 DPM performance tuning hints for HTTP/WebDAV and Xrootd

Emphasis  DPM emphasis has moved to better support Xrootd and HTTP/WebDav protocols w.r.t HEP historical data access and transfers ( e.g. SRM, RFIO)  Xrootd Federations are fundamental for some LHC experiments computing models  AAA for CMS and FAX for ATLAS  HTTP/Webdav based federation for LHCb is under evaluation and a prototype available:  /10/2014 DPM performance tuning hints for HTTP/WebDAV and Xrootd

RFIOD DPM HEAD NS) DPM HEAD NS) New Dmlite Architecture 7 16/10/2014 DPM performance tuning hints for HTTP/WebDAV and Xrootd DPM DPNS DMLITE MYSQL MEM CACHE Legacy SRM Grid FTP Grid FTP XRootD Web DAV Web DAV ADAPTER The goal is to reduce interactions towards legacy components ( bottleneck for performance)

Internal Performance Assessment 8 16/10/2014 DPM performance tuning hints for HTTP/WebDAV and Xrootd  For ~1 year we have been improving our sets of tests headed to METADATA performance and reliability under load  Goals are:  Perf tests: Regular maximum stress exercises on our internal testbeds  quick response: evaluate day by day any change in how the testbeds perform

DPM performance tuning hints for HTTP/WebDAV and Xrootd 9 16/10/2014  Nightly performance tests run towards the CERN dev testbeds  Very heavy stress generators  Protocols: HTTP, WebDav, Xrootd  Concurrency: sweeping from 1 to 1000  Tasks:  Peak metadata performance: repeated stat of 10k files by clients  Small reads from each file ( few bytes)  HW: 4 core, 8GB mem, 1GB Net Internal Performance Assessment

DPM performance tuning hints for HTTP/WebDAV and Xrootd 10 16/10/2014 Performance Results: release WebDav peak stat performance

DPM performance tuning hints for HTTP/WebDAV and Xrootd 11 16/10/2014 Performance Results: release HTTP small file GET peak performance

DPM performance tuning hints for HTTP/WebDAV and Xrootd 12 16/10/2014 Performance Results: release Xrootd stat rate

The CMS AAA tests 13 16/10/2014 DPM performance tuning hints for HTTP/WebDAV and Xrootd  CMS has started a performance assessment campaign to understand if sites can sustain the expected load for LHC Run2  Thanks to F. Fanzago for the info and the graphs!  CMS target for tests:  File-opening test: access total rate of 100 Hz at a site  Very aggressive test, as the entire CMS infra would access a single storage element  material/slides/1.pdf material/slides/1.pdf

– Tests run up 100 jobs simultaneously, opening files at rate of 2 Hz each. TEST TARGET IS 100 Hz Opening Plots for some DPM sites DPM performance tuning hints for HTTP/WebDAV and Xrootd 14 Federica Fanzago - INFN Padova

Performance Tuning DPM performance tuning hints for HTTP/WebDAV and Xrootd 15 16/10/2014  The internal performance tests + interaction with CMS sites with low performance has highlighted many performance bottlenecks and configuration parameters to be tuned  The new release is the result of this activity  Caching to avoid DPNS/DPM services interaction  WIKI for Performance Tuning valid for too  dmin/TuningHints dmin/TuningHints

Tuning Hints  Some tuning hints are related to Mysql connections and pool size  max connection=1000  NsPoolSize= 256  Others to number of threads for the legacy daemons:  DPM-> NB_FTHREADS=60  DPNS-> NB_THREADS=80  Limits also are important  hard nofile  soft nproc  soft nofile  hard nproc DPM performance tuning hints for HTTP/WebDAV and Xrootd 16 16/10/2014

Tuning Hints: Memcache  Usage of the Memcache plugin with a local instance or external cluster of Memcached server(s).  2GB cache with 256 connections pool size DPM performance tuning hints for HTTP/WebDAV and Xrootd 17 16/10/2014 DPM HEAD NS) DPM HEAD NS) DMLITE MYSQL MEM CACHE Grid FTP Grid FTP XRootD Web DAV Web DAV ADAPTER Mem cached Mem cached Mem cached Mem cached Mem cached Mem cached

Performance Tuning: Before tuning After tuning Strong collaboration between site manager and DPM developers Optimization of DPM configuration parameters Performance not optimal due to headnode’s hardware 18 Federica Fanzago - INFN Padova

DPM performance tuning hints for HTTP/WebDAV and Xrootd 19 16/10/2014 Performance Tuning: WebDav peak performance in … a 5X increase, no more clogging

DPM performance tuning hints for HTTP/WebDAV and Xrootd 20 16/10/2014 Performance Tuning: HTTP small file GET peak performance in 1.8.9, 20X increase

DPM performance tuning hints for HTTP/WebDAV and Xrootd 21 16/10/2014 Performance Tuning: Xrootd stat rate with DPM 1.8.9, 4X increase

Performance Tuning: Puppet  Configuration changes are now possible also with Puppet as it has become the official DPM configuration mechanism ( Yaim is deprecated)  We have been developing a series of modules to configure the various plugins and frontends  Available on PuppetForge and GitHub   DPM performance tuning hints for HTTP/WebDAV and Xrootd 22 16/10/2014

Summary  Good synergy between Experiments, Site Managers and DPM Developers  DPM Xrootd and HTTP/WebDav Federations are ready for Run2  DPM might satisfy the modern perf requirements, but needs powerful hardware  DPM should meet them with more normal machines, e.g. 4cores, 8G mem, good disks, average network /10/2014 DPM performance tuning hints for HTTP/WebDAV and Xrootd

DPM QUESTIONS? /10/2014DPM performance tuning hints for HTTP/WebDAV and Xrootd