Resource Predictors in HEP Applications John Huth, Harvard Sebastian Grinstein, Harvard Peter Hurst, Harvard Jennifer M. Schopf, ANL/NeSC.

Slides:



Advertisements
Similar presentations
Storing Data: Disks and Files: Chapter 9
Advertisements

Nondeterministic Queries in a Relational Grid Information Service Peter A. Dinda Dong Lu Prescience Lab Department of Computer Science Northwestern University.
Live Migration of Virtual Machines Christopher Clark, Keir Fraser, Steven Hand, Jacob Gorm Hansen, Eric Jul, Christian Limpach, Ian Pratt, Andrew Warfield.
Esma Yildirim Department of Computer Engineering Fatih University Istanbul, Turkey DATACLOUD 2013.
Active and Accelerated Learning of Cost Models for Optimizing Scientific Applications Piyush Shivam, Shivnath Babu, Jeffrey Chase Duke University.
Helper Threads via Virtual Multithreading on an experimental Itanium 2 processor platform. Perry H Wang et. Al.
Database Management Systems, R. Ramakrishnan and J. Gehrke1 External Sorting Chapter 11.
Yoshiharu Ishikawa (Nagoya University) Yoji Machida (University of Tsukuba) Hiroyuki Kitagawa (University of Tsukuba) A Dynamic Mobility Histogram Construction.
Benchmarking Parallel Code. Benchmarking2 What are the performance characteristics of a parallel code? What should be measured?
SKELETON BASED PERFORMANCE PREDICTION ON SHARED NETWORKS Sukhdeep Sodhi Microsoft Corp Jaspal Subhlok University of Houston.
Parallelized variational EM for Latent Dirichlet Allocation: An experimental evaluation of speed and scalability Ramesh Nallapati, William Cohen and John.
Predicting The Performance Of Virtual Machine Migration Presented by : Eli Nazarov Sherif Akoush, Ripduman Sohan, Andrew W.Moore, Andy Hopper University.
1 COMP 206: Computer Architecture and Implementation Montek Singh Mon., Sep 5, 2005 Lecture 2.
Data Management for Physics Analysis in PHENIX (BNL, RHIC) Evaluation of Grid architecture components in PHENIX context Barbara Jacak, Roy Lacey, Saskia.
UC Berkeley 1 Time dilation in RAMP Zhangxi Tan and David Patterson Computer Science Division UC Berkeley.
FALL 2006CENG 351 Data Management and File Structures1 External Sorting.
Present by Chen, Ting-Wei Adaptive Task Checkpointing and Replication: Toward Efficient Fault-Tolerant Grids Maria Chtepen, Filip H.A. Claeys, Bart Dhoedt,
1 stdchk : A Checkpoint Storage System for Desktop Grid Computing Matei Ripeanu – UBC Sudharshan S. Vazhkudai – ORNL Abdullah Gharaibeh – UBC The University.
1 External Sorting for Query Processing Yanlei Diao UMass Amherst Feb 27, 2007 Slides Courtesy of R. Ramakrishnan and J. Gehrke.
Peter Dinda Department of Computer Science Northwestern University Beth Plale Department.
Information and Scheduling: What's available and how does it change Jennifer M. Schopf Argonne National Lab.
Evaluating current processors performance and machines stability R. Esposito 2, P. Mastroserio 2, F. Taurino 2,1, G. Tortone 2 1 INFM, Sez. di Napoli,
Test results Test definition (1) Istituto Nazionale di Fisica Nucleare, Sezione di Roma; (2) Istituto Nazionale di Fisica Nucleare, Sezione di Bologna.
Accelerating SQL Database Operations on a GPU with CUDA Peter Bakkum & Kevin Skadron The University of Virginia GPGPU-3 Presentation March 14, 2010.
Edge Based Cloud Computing as a Feasible Network Paradigm(1/27) Edge-Based Cloud Computing as a Feasible Network Paradigm Joe Elizondo and Sam Palmer.
EPICS Archiving Appliance Test at ESS
Database Systems: Design, Implementation, and Management Eighth Edition Chapter 10 Database Performance Tuning and Query Optimization.
CMS Report – GridPP Collaboration Meeting VI Peter Hobson, Brunel University30/1/2003 CMS Status and Plans Progress towards GridPP milestones Workload.
Profiling Grid Data Transfer Protocols and Servers George Kola, Tevfik Kosar and Miron Livny University of Wisconsin-Madison USA.
03/27/2003CHEP20031 Remote Operation of a Monte Carlo Production Farm Using Globus Dirk Hufnagel, Teela Pulliam, Thomas Allmendinger, Klaus Honscheid (Ohio.
20 October 2006Workflow Optimization in Distributed Environments Dynamic Workflow Management Using Performance Data David W. Walker, Yan Huang, Omer F.
Data transfer over the wide area network with a large round trip time H. Matsunaga, T. Isobe, T. Mashimo, H. Sakamoto, I. Ueda International Center for.
Event Data History David Adams BNL Atlas Software Week December 2001.
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.
Building a Real Workflow Thursday morning, 9:00 am Greg Thain University of Wisconsin - Madison.
High-speed TCP  FAST TCP: motivation, architecture, algorithms, performance (by Cheng Jin, David X. Wei and Steven H. Low)  Modifying TCP's Congestion.
Analysis of the ROOT Persistence I/O Memory Footprint in LHCb Ivan Valenčík Supervisor Markus Frank 19 th September 2012.
1 Overview of IEPM-BW - Bandwidth Testing of Bulk Data Transfer Tools Connie Logg & Les Cottrell – SLAC/Stanford University Presented at the Internet 2.
Performance Model for Parallel Matrix Multiplication with Dryad: Dataflow Graph Runtime Hui Li School of Informatics and Computing Indiana University 11/1/2012.
Advisor: Resource Selection 11/15/2007 Nick Trebon University of Chicago.
Fast Crash Recovery in RAMCloud. Motivation The role of DRAM has been increasing – Facebook used 150TB of DRAM For 200TB of disk storage However, there.
ROOT and Federated Data Stores What Features We Would Like Fons Rademakers CERN CC-IN2P3, Nov, 2011, Lyon, France.
Computer Component. A computer is a machine that is used to store and process data electronically Computer Definition.
Active Sampling for Accelerated Learning of Performance Models Piyush Shivam, Shivnath Babu, Jeff Chase Duke University.
Replicating Memory Behavior for Performance Skeletons Aditya Toomula PC-Doctor Inc. Reno, NV Jaspal Subhlok University of Houston Houston, TX By.
U N I V E R S I T Y O F S O U T H F L O R I D A Hadoop Alternative The Hadoop Alternative Larry Moore 1, Zach Fadika 2, Dr. Madhusudhan Govindaraju 2 1.
PHENIX and the data grid >400 collaborators 3 continents + Israel +Brazil 100’s of TB of data per year Complex data with multiple disparate physics goals.
CMS Computing Model Simulation Stephen Gowdy/FNAL 30th April 2015CMS Computing Model Simulation1.
Preliminary Validation of MonacSim Youhei Morita *) KEK Computing Research Center *) on leave to CERN IT/ASD.
PROOF tests at BNL Sergey Panitkin, Robert Petkus, Ofer Rind BNL May 28, 2008 Ann Arbor, MI.
Introduction to Database Systems1 External Sorting Query Processing: Topic 0.
David Adams ATLAS Datasets for the Grid and for ATLAS David Adams BNL September 24, 2003 ATLAS Software Workshop Database Session CERN.
Sunpyo Hong, Hyesoon Kim
BNL dCache Status and Plan CHEP07: September 2-7, 2007 Zhenping (Jane) Liu for the BNL RACF Storage Group.
An Analysis of Memory Access in Relation to Operations on Data Structures Ryan Connaughton & Daniel Rinzler CSE December 13, 2006 An Analysis of.
Operating Systems A Biswas, Dept. of Information Technology.
18 May 2006CCGrid2006 Dynamic Workflow Management Using Performance Data Lican Huang, David W. Walker, Yan Huang, and Omer F. Rana Cardiff School of Computer.
CS 540 Database Management Systems
CS 704 Advanced Computer Architecture
Lecture 2: Performance Evaluation
U.S. ATLAS Grid Production Experience
Solid State Disks Testing with PROOF
Optimizing the Migration of Virtual Computers
David Front Weizmann institute May 2007
Memory Management for Scalable Web Data Servers
Grid Canada Testbed using HEP applications
Predictive Performance
Alice Software Demonstration
CENG 351 Data Management and File Structures
Presentation transcript:

Resource Predictors in HEP Applications John Huth, Harvard Sebastian Grinstein, Harvard Peter Hurst, Harvard Jennifer M. Schopf, ANL/NeSC

The Problem Large data sets gets recreated, and scientists want to know if they should –Fetch a copy of the data –Recreate it locally This problem can be considered in the context of a virtual data system that tracks how data is created so recreation is feasible

To make this decision you need 1) Estimate of time to recreate data –Info about data provenance, machine types, etc 2) Estimate of data transfer time 3) Framework to allow you to take advantage of these choices by adapting the workflow accordingly

To make this decision you need 1) Estimate of time to recreate data –Info about data provenance, machine types, etc 2) Estimate of data transfer time 3) Framework to allow you to take advantage of these choices by adapting the workflow accordingly –OUR AREA OF CONCENTRATION

Regeneration Time Estimates Previous work (Chep 2004, “Resource Predictors in HEP Applications”) Estimate runtime of ATLAS application –End-to-end estimation since no low-level application model available –Used data about input parameters (number of events, versioning, debug on/off, etc) and benchmark data (using nbench) Estimates are accurate to 10% for event generation and reconstruction, 25% for event simulation

Regeneration Time Estimate Accuracy

File Transfer Time Estimates Much previous work (e.g. Vazhkudai and Schopf, IJHPCA Vol 17, No. 3, August 2003 ) We use simple end-to-end history data from GridFTP logs to estimate behavior –Simple approach works well on our networks/machines –Average bandwidth used with no file-size filtering

Testbed Size of file transferred (was it the same all the time or did it vary?) Three sites – Boston, Cern, BNL Something about the machines – esp OS, arch, and memory size Something about the standard latency Something about the network – what was the bottleneck in terms of smallest piece – 100 MBS ethernet to something? Something about the variance seen

Testbed Files transferred from BNL to Harvard and from CERN to Harvard –BNL (aftpexp01.bnl.gov): 4x 3GHz Xeon, Linux ELsmp, 2.0GB RAM, 1.0 GBit/s NIC –Harvard: 2x 3.4GHz P4, Linux EL.cernsmp, 1.5GB RAM, 1.0 GBit/s NIC Typical network routes: –Harvard –NoX – ManLan – ESNet – BNL Typical Latency 7.8 ms –Harvard – NoX – ManLan – Chicago (Abilene) – CERN Typical Latency 148 ms Bottlenecks are in machines at each end (e.g. disk access)

Initial Transfer History Transferred 20 files each, 25MB, 50MB, 100MB, 250MB, 500MB, 1GB from BNL to Harvard (similar tests for CERN to Harvard link) – BNL: aftpexp01.bnl.gov (4x3.0GHz Xeon, Linux ELsmp, 2.0GB RAM, 1Gbit/sec NIC) – Harvard: heplatlas3.physics.harvard.edu (2x3.4GHz P4, Linux EL.cernsmp, 1.5GB RAM, 1Gbit/sec NIC) –CERN: castorgrid.cern.ch - Linux Typical routes: –Harvard – NoX – ManLan – ESNet – BNL Typical latency 7.8ms –Harvard – NoX – ManLan – Chicago (Abilene) – CERN Typical latency 148ms Bottlenecks are apparently within the internal Harvard, BNL networks. Harvard, BNL machines, network very quiet during this initial phase Transfer times are linear with file size During this quiet time, transfers of typical 100MB ATLAS files have typical variance of approximately 5%

Network Routing

Transfer Benchmarking Transfer files from BNL to Harvard –20 files each 25MB, 50MB, 100MB, 250MB, 500MB, 1GB Average file transfer times are linear with file size Initially quiet machines, network –Transfers of 100MB files have variance ~5%

Time vs File Size, BNL (Quiet network)

Transfer Variance, BNL (100 MB files, quiet network)

Transfer Benchmarking Some data taken during “Service Challenge 3” Average file transfer times are still linear with file size, but have larger variance

Time vs File Size, BNL (Busy network)

Transfer Variance, BNL (100 MB files, busy network)

But our concentration was on the framework Given ways to estimate application run time and file transfer time, we want to plug them into an existing framework to make better resource management decisions Could be implemented as a post- processor to optimize DAG’s produced by Chimera

Workflow Optimization A script parses the DAG, looking for I/O, binaries I/O files indexed in Replica Location Service (RLS) Client queries database for execution parameters, bandwidths Script evaluates execution, transfer times and rewrites fastest DAG

Workflow Optimization Steps A script parses job submission files, looks for I/O file names, event numbers, names of binaries Locate files in RLS, determine sizes, provenance Client queries database for execution time parameters and end-to-end bandwidths Go job-by-job –Calculate regeneration time estimate –Calculate file transfer time estimate –Rewrite job files to perform fastest instantiation

Our Strawman Application ATLAS event reconstruction jobs take ~20Mins to calculate a 100 Meg file File transfer Boston to BNL ~15 Sec/ 100 Meg file We created simplified jobs that would have average execution times equal to the file transfer times in order to have a situation closer to the one originally hypothesized Likely to be more common as data access becomes more contentious, and machines/calculations speed up

Framework Tests Generate “Non-optimized” DAG’s – linear chains which use a random mixture of transfers and calculations to instantiate 10, 20, or 40 files. Operate on these DAG’s with our optimizer to produce “Optimized” DAG’s Submit both “Non-optimized” and “Optimized” DAG’s and compare processing times For our particular strawman we expect the “Optimized” DAG’s to be 25% faster than the “Non-optimized”

Framework Tests

Comparison of Results

Optimized Results

Summary Implementation works A 28% time savings is seen Works with crude bandwidth predictions –More sophisticated predictions for dynamic situations would be helpful Most useful when regeneration and transfer times are similar.