Archive Systems What you always wanted to know

Archive Systems What you always wanted to know but were afraid to ask: What’s available? Who’s doing what? PAL EPICS Meeting Oct. 2012

1. Performance? How many samples/second write & read? 2

2. How to get at the data? • CSS archive plugin? • Easy web reports? 3

3. All the Rest? • How to configure? • How to – start/stop sample engines? – Monitor if they’re • running? • connected to all channels? • write without errors? • Impact of disk failure, engine crash? • Limits when data grows? • How to spread data across disks, servers? – How to delete certain channels or time ranges? – Data backup? • Info: – Which channel has the most samples in last 24 hours? – Which subsystem generates the most samples? 4

System • Language, Platform: Java, C++, ? ? • Data Store ü Advantage ü Great Stuff - Disadvantage - Problem User, user 6

Channel Archiver • C++ • Custom Data/Index Files ü Fast: Write 60 k, 70 k samples/sec - Need library to read data - Trouble when engine crashed - File size limits - Maintenance nightmare SNS until 2009, KEK, J-PARC, SLAC, KSTAR, . . 7

Channel Archiver, 64 bit. . As before, but ü Extended file size limit - Even Bigger Trouble when engine crashed? - Even Bigger Data maintenance nightmare? Diamond 8

RDB-Based Channel Archiver (BEAUTY) • Java (Eclipse, CSS) • RDB: My. SQL, Postgre. SQL, Oracle, … ü Any data maintenance possible ü No size limits; partitions, cluster, … ü Any type of reporting on data ü Anybody can reach data from any programming language ü Engines don’t crash, and if they do: Just no new samples for that engine - Slow: 8000 samples/sec write - ~15 k for special My. SQL setup SNS since 2009, KEK, ITER? 9

DESY My. SQL Archiver • Java (started with BEAUTY code) • RDB: My. SQL with special setup – Writing original sample plus a few averages ü Faster writing because of some My. SQL optimization ü Faster retrieval of long time ranges ü Most of the usual RDB benefits - Only My. SQL DESY? 10

Fermilab (not EPICS) • Java • My. SQL ‘BLOBs’ used as ring buffer ü Faster ü Automatically replace oldest data, not running out of space - Need library to write and read, encode and decode content of BLOBs - None of the usual RDB advantages Fermilab 11

My. A (Jlab) • C++? • Homebrew My. SQL ‘cluster’ – Several My. SQL Servers. One table per channel. ü Faster ü Most of the usual RDB benefits - Need library to read from correct My. SQL server and table JLab 12

Archive Appliance • Java • Google Protocol Buffers ü Faster ü “Appliance” Idea üSubmit channel name. Rest Auto-configured. - Need library to access data? SLAC, BNL, IHEP, PSI, ITER? 13

CERN • Language, Platform: Java, C++, ? ? • Oracle ü Fast because they tuned it well - not EPICS CERN 14

SDDS • C • SDDS Files ü Integrate 4 d w/ SDDS tools - Needs library to access - Data management APS, … 15

Fast Acqu. Archive • C • Raw disk ü Fast, for specific purpose - Need lib. - Data management Diamond 16

SLC History Buffers • Fortran • Binary files ü Worked for decades - Slow SLAC 17