ROOT IO workshop What relates to ATLAS. General Both CMS and Art pushing for parallelization at all levels. Not clear why as they are anyhow CPU bound.

Slides:



Advertisements
Similar presentations
WSN Simulation Template for OMNeT++
Advertisements

File System. NET+OS 6 File System Architecture Design Goals File System Layer Design Storage Services Layer Design RAM Services Layer Design Flash Services.
XAOD persistence considerations: size, versioning, schema evolution Joint US ATLAS Software and Computing / Physics Support Meeting Peter van Gemmeren.
Wahid Bhimji University of Edinburgh J. Cranshaw, P. van Gemmeren, D. Malon, R. D. Schaffer, and I. Vukotic On behalf of the ATLAS collaboration CHEP 2012.
ALICE data access WLCG data WG revival 4 October 2013.
ATLAS Peter van Gemmeren (ANL) for many. xAOD  Replaces AOD and DPD –Single, EDM – no T/P conversion when writing, but versioning Transient EDM is typedef.
 Optimization and usage of D3PD Ilija Vukotic CAF - PAF 19 April 2011 Lyon.
David N. Brown Lawrence Berkeley National Lab Representing the BaBar Collaboration The BaBar Mini  BaBar  BaBar’s Data Formats  Design of the Mini 
Costin Grigoras ALICE Offline. In the period of steady LHC operation, The Grid usage is constant and high and, as foreseen, is used for massive RAW and.
Introduction Advantages/ disadvantages Code examples Speed Summary Running on the AOD Analysis Platforms 1/11/2007 Andrew Mehta.
The Alternative Larry Moore. 5 Nodes and Variant Input File Sizes Hadoop Alternative.
Overview and Applications
Operating Systems Lecture November 2015© Copyright Virtual University of Pakistan 2 Agenda for Today Review of previous lecture Hardware (I/O, memory,
ROOT and Federated Data Stores What Features We Would Like Fons Rademakers CERN CC-IN2P3, Nov, 2011, Lyon, France.
CHEP ‘06 David Lawrence JLab C++ Introspection and Object Persistency through JIL David Lawrence Ph. D. Jefferson Lab, Newport News VA.
CHEP 2013, Amsterdam Reading ROOT files in a browser ROOT I/O IN JAVASCRIPT B. Bellenot, CERN, PH-SFT B. Linev, GSI, CS-EE.
Optimizing I/O Performance for ESD Analysis Misha Zynovyev, GSI (Darmstadt) ALICE Offline Week, October 28, 2009.
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. Concurrency & Dynamic Programming.
The JANA Reconstruction Framework David Lawrence - JLab May 25, /25/101JANA - Lawrence - CLAS12 Software Workshop.
Online Monitoring System at KLOE Alessandra Doria INFN - Napoli for the KLOE collaboration CHEP 2000 Padova, 7-11 February 2000 NAPOLI.
A. Gheata, ALICE offline week March 09 Status of the analysis framework.
ROOT Workshop M.Frank LHCb/CERN Improvements in the I/O Area (*)  General I/O related improvements  Tree related issues  Plans (*) I present.
Analysis experience at GSIAF Marian Ivanov. HEP data analysis ● Typical HEP data analysis (physic analysis, calibration, alignment) and any statistical.
Improving the analysis performance Andrei Gheata ALICE offline week 7 Nov 2013.
Analysis efficiency Andrei Gheata ALICE offline week 03 October 2012.
Lecture 4 Page 1 CS 111 Online Modularity and Memory Clearly, programs must have access to memory We need abstractions that give them the required access.
I/O aspects for parallel event processing frameworks Workshop on Concurrency in the many-Cores Era Peter van Gemmeren (Argonne/ATLAS)
 IO performance of ATLAS data formats Ilija Vukotic for ATLAS collaboration CHEP October 2010 Taipei.
PROOF on multi-core machines G. GANIS CERN / PH-SFT for the ROOT team Workshop on Parallelization and MultiCore technologies for LHC, CERN, April 2008.
Barthélémy von Haller CERN PH/AID For the ALICE Collaboration The ALICE data quality monitoring system.
Multi Process I/O Peter Van Gemmeren (Argonne National Laboratory (US))
ROOT : Outlook and Developments WLCG Jamboree Amsterdam June 2010 René Brun/CERN.
ANALYSIS TRAIN ON THE GRID Mihaela Gheata. AOD production train ◦ AOD production will be organized in a ‘train’ of tasks ◦ To maximize efficiency of full.
Peter van Gemmeren (ANL) Persistent Layout Studies Updates.
Efi.uchicago.edu ci.uchicago.edu Caching FAX accesses Ilija Vukotic ADC TIM - Chicago October 28, 2014.
Deadlock Avoidance Determine whether, by allowing allocation, we could get a deadlock, i.e. check if by allowing allocation the system could enter a state.
The heavyweight parts of lightweight languages LL1 Workshop November 17, 2001.
Data Formats and Impact on Federated Access
Atlas IO improvements and Future prospects
Chapter 2: Computer-System Structures(Hardware)
Chapter 2: Computer-System Structures
ALICE experience with ROOT I/O
Why Events Are A Bad Idea (for high-concurrency servers)
Component 2 6G, H, I, J, K.
Fibers – blocking is cheap in a Parallel Universe jPrime Stefan Minev
Input/Output and Communication
Multicore Computing in ATLAS
Spark Presentation.
The heavyweight parts of lightweight languages
10.3 Details of Recursion.
Sharing Memory: A Kernel Approach AA meeting, March ‘09 High Performance Computing for High Energy Physics Vincenzo Innocente July 20, 2018 V.I. --
2018 I/O POW Ph. Canal, B. Bockelman, D. Piparo, Jakob B., J. Pivarski, Z. Zhang, O. Shadura.
Comments about PROOF / ROOT evolution during coming years
Analysis framework - status
Modularity and Memory Clearly, programs must have access to memory
ALICE Computing Model in Run3
Silberschatz, Galvin and Gagne  Operating System Concepts Chapter 2: Computer-System Structures Computer System Operation I/O Structure Storage.
Chapter 8: Main Memory.
Chapter 9: Virtual-Memory Management
Communication and Memory Efficient Parallel Decision Tree Construction
Security in Java Real or Decaf? cs205: engineering software
Chapter 8: Memory management
Outline Module 1 and 2 dealt with processes, scheduling and synchronization Next two modules will deal with memory and storage Processes require data to.
Operating Systems Lecture 3.
Database Design and Programming
Program Design Language (PDL)
Concurrency, Processes and Threads
Chapter 2: Computer-System Structures
Chapter 2: Computer-System Structures
Summary of the dCache workshop
Presentation transcript:

ROOT IO workshop What relates to ATLAS

General Both CMS and Art pushing for parallelization at all levels. Not clear why as they are anyhow CPU bound and simple forking will spare them more than enough memory Art wants native byte ordering, C++14 support, lightweight, thread safe histogram object Alice now goes the way of ATLAS AOD flattening. Requests asynchronous de-serialization, thread safe hisotgrams and trees without locking.

ATLAS We want to ensure xAODs are WAN access efficient. as small as possible need smallest possible number of read requests have it all in root even if we can have it in our framework

Requests Old request – TTreeCache to be ON by default – coded but still in testing by David Smith. Compression – ESDs and AODs had custom compression algorithms in transient/persistent layer of IO. That is gone with xAODs. Size of some collections went up a lot. Need to recuperate at least a part. – Two main custom compressions: stripping bits (reduces precision) and covariance matrix compression – ROOT provides float16 and double32 types but that’s not sufficient and is very difficult to use – We requested new methods for creating branches, where one could specify precision desired. – ROOT team agreed.

Requests Optimized basket size – Better algorithm written 7 years ago. Tested on ATLAS data (not xAOD) and CMS. – Reduces total number baskets in file, improves occupancy – With less reads needed improves IO rates, slightly reduces data size, memory consumption. – Issues explained to David Smith. He will work on it.

Error handling We have all seen cases where ROOT does not handle well errors on accessing xrootd files, specially when using TChain. – ROOT team stance – report as many bug reports as needed. – Will not throw exceptions, but will correctly forward error codes. – Not a trivial thing to find/reproduce a problem as it can happen at any moment of ROOT/xrootd plugin interaction.

AsynchronousPrefetch Our current xAOD IO libraries by default use 20MB of TTC, but no AsyncPrefetch Change in how implemented (compared to what explained at last CHEP). Now equivalent as adding a stream. This could mean up to factor 2 read speed gain. We must extensively test this.

Monitoring xAOD IO Discussed between Attila, Paul, Ilija All the data accesses will be monitored (both Athena and straight ROOT) Initial list of variables agreed upon Initial code collecting the information should be checked in today Info will be sent to the collection system Ilija will provide collecting and mining part, now coding different variants (MongoDB, GAE, Oracle). When it proves itself in ATLAS, ROOT people would really like to have the system monitor as large part of all the ROOT jobs as possible. Quite different set of variables to collect. ROOT team will try to find someone from CERN IT to instrument ROOT.