Testing High Performance Tape Drives HEPi X FALL

Testing High Performance Tape Drives HEPi. X FALL 2005 Data Services Section Hugo Caçote @ HEPi. X Fall 2005

Motivation LHC 7 0 0 2 ~15 Petabytes/year Hugo Caçote @ HEPi. X Fall 2005

Current Model DATA RFIO DATA CASTOR HSM DATA Hugo Caçote @ HEPi. X Fall 2005

Devices Tested IBM 3592 JA Tape Drive IBM 3584 Library + + 300 GB / 60 GB 40 MB/sec Future Generation Tape Drives >100 MB/s 117 carts 12 drives Testing in 2006 HP LT 0 -3 + STK SL 8500 + 400 GB 1, 448 carts 64 drives 80 MB/sec Hugo Caçote @ HEPi. X Fall 2005

Test Infrastructure Tape Server Tape Library Fiber Channel HBA Tape Drive Hugo Caçote @ HEPi. X Fall 2005

Functionality Tests • Go through the set of commands available on the SCSI standard Check returned information, timing, command acceptance SCSI COMMANDS: Change Definition, Compare, Copy and Verify, Display Message, Erase, Format Medium, Inquiry, Load/Unload, Locate , Log Select, Log Sense, Mode Select (6), Mode Select (10), Mode Sense (6), Mode Sense (10), Persistent Reserve In, Persistent Reserve Out, Prevent/Allow Medium Removal, Read Attribute, Read Block Limits, Read Buffer, Read Position, Read Reverse, Receive Diagnostic Results, Recover Buffered Data, Release Unit (6), Release Unit (10), Report Density Support LUNs, Request Sense Unit (6), Reserve Unit (10), Rewind, Send Diagnostic, Set Capacity, Space, Test Unit Ready, Verify, Write Attribute, Write Buffer, Write Filemarks Hugo Caçote @ HEPi. X Fall 2005

Functionality Tests fibre channel analyzer for verifying SCSI commands Hugo Caçote @ HEPi. X Fall 2005

Test Scenarios Linux tape driver and cern. Tape. Test. Util (interactive/command line mode) Hugo Caçote @ HEPi. X Fall 2005

Analysis of Results Hugo Caçote @ HEPi. X Fall 2005

Mechanical Tests IBM 3592: Over 125, 000 mount/dismount cycles performed, no errors Test mechanical reliability of drive / media: some cartridges now mounted > 4000 times, no errors Random file reads on selected tapes and media: superseded by CASTOR operation in data challenges HP LTO-3: Over 125, 000 mount/dismount cycles performed, no errors Test mechanical reliability of drive / media: some cartridges now mounted > 5000 times, no errors Hugo Caçote @ HEPi. X Fall 2005

Performance Tests - Use of native Linux Commands (mt/dd) for data transfers : • • • read / write compression / no compression blocksize filesize position “labeled” files Hugo Caçote @ HEPi. X Fall 2005

Performance LTO-3 Data Transfer Rate Write no compression Hugo Caçote @ HEPi. X Fall 2005

Performance LTO-3 Data Transfer Rate Read no compression Hugo Caçote @ HEPi. X Fall 2005

Linear Serpentine Recording Technique Hugo Caçote @ HEPi. X Fall 2005

Performance LTO-3 Locate File Timing LTO-3 Locate Record Timing r lde o n e se ) (? (? ) m P e l ob st H r u p m es, j h rit iv go 1 dr l A OLT END OF TAPE BEGINING OF TAPE Hugo Caçote @ HEPi. X Fall 2005

ANSI Labels 80 Bytes Header 1 Header 2 Header 3 0 -? Bytes Tapemark Data Tapemark 80 Bytes Trailer 1 Trailer 2 Trailer 3 Headers Filename, block size, HSM version, time of writing … Trailers Number of blocks, non standard data … Tapemark Sync Tapemark Special records on tape used by the drive , immediate bit =0/1 Flush buffer Hugo Caçote @ HEPi. X Fall 2005 Sync

Labels vs Performance Minimum over head Maximum over head Hugo Caçote @ HEPi. X Fall 2005

Labels vs Performance HP-LTO 3 IBM 3592 JA Hugo Caçote @ HEPi. X Fall 2005

Labels vs Performance IBM 3592 small files Hugo Caçote @ HEPi. X Fall 2005

HSM Integration Functionality + Mechanical + Performance OK Drive integration in HSM system tape_up tape unit standard testing for production utility tplabel tape labelling utility dumptape dumping (scanning) utility stagein tape reading utility stagewrt tape writing utility repack move CASTOR file from tape and reclaim utilities Hugo Caçote @ HEPi. X Fall 2005

Operations IBM 3592/HP LT 0 -3: SNMP Agent : Error Counters Tape Alerts Drive and Media: number mounts/Loads/… IBM 3592: Statistical Analysis and Reporting System : bit 62: SARS Drive Relative Quality X'00' is unknown, best X'01' -> worst X'FF‘ Request Sense bit 63 SARS Media Relative Quality X'00' is unknown, best X'01' -> worst X'FF' No Vendor: drives Perfect tool for monitoring all type of drives and for a large number of Hugo Caçote @ HEPi. X Fall 2005

More Tests RFIO CASTOR HSM ? ? SAN $$ Hugo Caçote @ HEPi. X Fall 2005