Working Items and plan Hideki Miyake (KEK) June 30 th, 2014 KISTI, Korea.

Slides:



Advertisements
Similar presentations
Maximizing COM+ Application Throughput Steve Swartz Program Manager COM+ Team Microsoft Corporation
Advertisements

Lecture 36: Chapter 6 Today’s topic –RAID 1. RAID Redundant Array of Inexpensive (Independent) Disks –Use multiple smaller disks (c.f. one large disk)
SEDA: An Architecture for Well-Conditioned, Scalable Internet Services Matt Welsh, David Culler, and Eric Brewer Computer Science Division University of.
File System Implementation
GGF Toronto Spitfire A Relational DB Service for the Grid Peter Z. Kunszt European DataGrid Data Management CERN Database Group.
Precept 3 COS 461. Concurrency is Useful Multi Processor/Core Multiple Inputs Don’t wait on slow devices.
Google Bigtable A Distributed Storage System for Structured Data Hadi Salimi, Distributed Systems Laboratory, School of Computer Engineering, Iran University.
CSE 451: Operating Systems Winter 2010 Module 13 Redundant Arrays of Inexpensive Disks (RAID) and OS structure Mark Zbikowski Gary Kimura.
National Manager Database Services
Distributed File Systems Concepts & Overview. Goals and Criteria Goal: present to a user a coherent, efficient, and manageable system for long-term data.
PMIT-6102 Advanced Database Systems
Lecture On Database Analysis and Design By- Jesmin Akhter Lecturer, IIT, Jahangirnagar University.
Distributed Systems. Interprocess Communication (IPC) Processes are either independent or cooperating – Threads provide a gray area – Cooperating processes.
Physical Database Design & Performance. Optimizing for Query Performance For DBs with high retrieval traffic as compared to maintenance traffic, optimizing.
Orbited Scaling Bi-directional web applications A presentation by Michael Carter
1 CMSC 341 Extensible Hashing Chapter 5, Section 6 (pp. 200 – 203)
1 CS 430 Database Theory Winter 2005 Lecture 16: Inside a DBMS.
Frontiers in Massive Data Analysis Chapter 3.  Difficult to include data from multiple sources  Each organization develops a unique way of representing.
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.
Enabling Grids for E-sciencE EGEE-III INFSO-RI I. AMGA Overview What is AMGA Metadata Catalogue of EGEE’s gLite 3.1 Middleware Main Feature of.
The Advanced Data Searching System The Advanced Data Searching System with 24 February APCTP 2010 J.H Kim & S. I Ahn & K. Cho on behalf of the Belle-II.
Database Replication in Tashkent CSEP 545 Transaction Processing Sameh Elnikety.
ESRI User Conference 2004 ArcSDE. Some Nuggets Setup Performance Distribution Geodatabase History.
Unified scripts ● Currently they are composed of a main shell script and a few auxiliary ones that handle mostly the local differences. ● Local scripts.
DELETION SERVICE ISSUES ADC Development meeting
08-Nov Database TEG workshop, Nov 2011 ATLAS Oracle database applications and plans for use of the Oracle 11g enhancements Gancho Dimitrov.
H IGH E NERGY A CCELERATOR R ESEARCH O RGANIZATION KEKKEK High Availability iRODS System (HAIRS) Yutaka Kawai, KEK Adil Hasan, ULiv December 2nd, 20091Interoperability.
Silberschatz, Galvin and Gagne ©2009 Operating System Concepts – 8 th Edition File System Implementation.
Measuring the Capacity of a Web Server USENIX Sympo. on Internet Tech. and Sys. ‘ Koo-Min Ahn.
HADOOP DISTRIBUTED FILE SYSTEM HDFS Reliability Based on “The Hadoop Distributed File System” K. Shvachko et al., MSST 2010 Michael Tsitrin 26/05/13.
Grid Technology CERN IT Department CH-1211 Geneva 23 Switzerland t DBCF GT Upcoming Features and Roadmap Ricardo Rocha ( on behalf of the.
Code Development for High Performance Servers Topics Multithreaded Servers Event Driven Servers Example - Game Server code (Quake) A parallelization exercise.
AFS/OSD Project R.Belloni, L.Giammarino, A.Maslennikov, G.Palumbo, H.Reuter, R.Toebbicke.
Sorting Criteria and more… Subtask 5 – RSD Robot System Design.
Tackling I/O Issues 1 David Race 16 March 2010.
StoRM + Lustre Proposal YAN Tian On behalf of Distributed Computing Group
Core and Framework DIRAC Workshop October Marseille.
ECE 297 Concurrent Servers Process, fork & threads ECE 297.
Bigtable A Distributed Storage System for Structured Data.
Performance Tuning Renegade
Jean-Philippe Baud, IT-GD, CERN November 2007
CS 540 Database Management Systems
Threads vs. Events SEDA – An Event Model 5204 – Operating Systems.
Security and Replication of Metadata with AMGA
UPC Parallel I/O Library
Lecture 12 Virtual Memory.
Chilimbi, et al. (2014) Microsoft Research
Enrico Fattibene CDG – CNAF 18/09/2017
Summary Set up testing environment on KISTI AMGA server
CSE-291 (Cloud Computing) Fall 2016
Objectives Solve compound inequalities with one variable.
File System Implementation
Conditions Data access using FroNTier Squid cache Server
Database Performance Tuning and Query Optimization
DUCKS – Distributed User-mode Chirp-Knowledgeable Server
-A File System for Lots of Tiny Files
SETL: Efficient Spark ETL on Hadoop
CSE 451: Operating Systems Winter 2009 Module 13 Redundant Arrays of Inexpensive Disks (RAID) and OS structure Mark Zbikowski Gary Kimura 1.
Half-Sync/Half-Async (HSHA) and Leader/Followers (LF) Patterns
Mark Zbikowski and Gary Kimura
CSE 451: Operating Systems Winter 2012 Redundant Arrays of Inexpensive Disks (RAID) and OS structure Mark Zbikowski Gary Kimura 1.
Design Components are Code Components
Overall Kernel Module Design
Chapter 2: Operating-System Structures
Virtual University of Pakistan
DISTRIBUTED SYSTEMS Principles and Paradigms Second Edition ANDREW S
Introduction To Distributed Systems
CMSC 341 Extensible Hashing.
Chapter 2: Operating-System Structures
Concurrent Cache-Oblivious B-trees Using Transactional Memory
Presentation transcript:

Working Items and plan Hideki Miyake (KEK) June 30 th, 2014 KISTI, Korea

The list shown in May AMGA topics related to Belle II - replication doesn't work - to avoid table collisions - stuck due to deep directory structure (below /belle/user/wenjing) - too many sessions for one AMGA operation - due to disconnect() - Can GSI socket keep sessions? --> significant improve - Can increase number of FE process (i.e. amgad) - connection pool? - Can decrease number of BE process (e.g. AMGA threading mode)? - Should check current I/O load which limits BE performance - performance against accesses during very short period - an API to scan AMGA directories - AMGA server was down by memory piled up - worry about future massive accesses - performance estimation - Initial goal: 10Hz (>30K jobs/1h duration) writing+deleting? - Should estimate more reasonable goal (production+analysis) - 1 second access time per directory is not acceptable - any idea to reduce access time? - synchronization between FC + metadata - relevant to AMGA survival model (LFC+AMGA, DFC+AMGA...) - future support - major concern from Belle II collaborators - AMGA development plan independent from Belle II? - AMGA development plan dedicated for Belle II? - activity invisibility from Belle II 2 Recent Belle II topics2014/6/30

The list shown in May AMGA topics related to Belle II - replication doesn't work - to avoid table collisions - stuck due to deep directory structure (below /belle/user/wenjing) - too many sessions for one AMGA operation - due to disconnect() - Can GSI socket keep sessions? --> significant improve - Can increase number of FE process (i.e. amgad) - connection pool? - Can decrease number of BE process (e.g. AMGA threading mode)? - Should check current I/O load which limits BE performance - performance against accesses during very short period - an API to scan AMGA directories - AMGA server was down by memory piled up - worry about future massive accesses - performance estimation - Initial goal: 10Hz (>30K jobs/1h duration) writing+deleting? - Should estimate more reasonable goal (production+analysis) - 1 second access time per directory is not acceptable - any idea to reduce access time? - synchronization between FC + metadata - relevant to AMGA survival model (LFC+AMGA, DFC+AMGA...) - future support - major concern from Belle II collaborators - AMGA development plan independent from Belle II? - AMGA development plan dedicated for Belle II? - activity invisibility from Belle II 3 Recent Belle II topics2014/6/30 Evaluated Good sign! Agreed tentative goal 100 operations/sec for both reading and writing

Proposal for this week First of all we should classify the items for short term (in this week) and others Short and easy items should be solved during the week –Want to have session to work together –Show our issue –Tackle the issue For the others we should determine rough plan and milestone (which should be shown in any Belle II meeting) 4 Recent Belle II topics2014/6/30

Memo Writing Limited number of concurrent access GSI patch significantly improved the performance since # session is dramatically decreased as expected (can drop disconnect()!!) Scalability issue still exists (# jobs  100K) AMGA proxy (DIRAC service) AMGA threading (multiple AMGA vs one DB) Access speed (O(1)) Reading Access speed (O(1)) for one directory  request two order faster Tricky API to read the directory

Memo Metadata Practical use case Attribute migration? Backend robustness Connection pool? Possibility of DFC+AMGA proposal What is advantage compared with DFC

Memo Development plan for AMGA server Development plan for AMGA client Tuning suggestion for PGSQL Possibility of MySQL? AMGA separation (federation) /belle/data /belle/MC /belle/user (or all others)

Memo Replication failed due to complicated directory Correct procedure Modification? Multi layered dataset Possible? Any suggestion

Open Question The lowest cost to check connectivity? “whoami” looks a bit complicated… Simple way like ping? Variable array? Or fixed array example? Best tuning parameter for PostgreSQL Backend possibility of MySQL? DB expansion (when records exceed disk size) 2014/6/30Recent Belle II topics 9

Backend benchmark Check transaction performance (not AMGA but backend DB) pgbench –t 5 (average of 5 measurements) Recent Belle II topics 10 Peak processing power < 100 connections? Most important thing is to handle a few thousands connections without error But performance optimization should be also important Is there room to optimize? Need your help! Limited by DISK I/O? Transaction / sec (TPS) # of clients Backend benchmark 2014/6/30

Possible solutions Recent Belle II topics 11 AMGApgpool DB1 DB2 replication AMGA1 AMGA2 - Connection pooling + multiple backends - AMGA federation DB1 DB2 -DB1 and DB2 are independent pooling >1000 ~100 /belle/MC /belle/data 2014/6/30