Hello to ALL;
this post is addressed to ALL MAXDB-GURU's !!
I'v few questions to MAXDB Performance-Guy's for R3load Export on MAXDB!
(remark to my person:
I'm certified OS/DB migration consultant and have already done over 500 OS/DB migrations since 1995 successfully)
BUT now I'm face with following setup for Export:
HP- IA64 BL870c, 4 CPUu2019s Dual-Core + Hyperthreading activ (mean's 16 "CPU'S" shown via top or glance)
64 GB RAM, HPUX V11.31;
MAXDB 7.7.07
ECC 5.0; Basis/Aba 640.22
SAP-Kernel 640-Unicode Patch 381
8 sapdatas are configured on unique 200GB LUN's and VG's in HP-UX; on storage-side the 8 220GB LUN's are located on (only) 9 300GB-Disks with Raid-5 Level and within HP-UX each VG/LVOL is mirrored via LVM to a second desasterdatacenter (200m distance)
LOGFILES:
2 x 4 GB LUN Raid-1 on Storage and via LVM also mirrored to failover-center
MAXDB-Datasize: 1600 GB Overall and within this 1350 GB used, TEMPUsage about 25 GB !
MAXDB-Parameter-Settings
MAXCPUu2019s 10 (4 x IA64 QuadCore+Hyperthreading shows 16 Cores within top and galcne
)
Cache_SIZE= I/O Buffer Cache = 16 GB (2.000.000 Pages)
Data-Cache-HitRatio: 99.61%
Undo-Cache = 99,98%
the following sapnote for MAXDB Peformance and Migrations are well known and already processes
928037, 1077887, 1385089, 1327874, 954268, 1464560, 869267,
My major problem is the export-runtime with over 4 days on the first dry-run (6 R3load's running on Linux-APPL-Server), and 46h on second runtime, 6 R3loads running on DB-Server IA64)
the third trail-run was aborted by me after 48hours and 50% export-dump-space was written. In all 3 dry-runs, no more than approx 3.5GB DUMP-Space were written per hour!
My first question to all MAXDB-Guru'S: How can I influence/optimize the TEMP - Area in MAXDB?? I didn't find any hint's in SDN or SAPNOTES or MaxDB Wiki or google....As fare as I know, this TEMP area "resides" within MAXDB-datafiles, thus it's seperated on my 48 datafiles, spreaded over 8 LUN/VG/disks. But I see LESS throughput on disk's and MAXDB-Kernel uses only ONE of ten's cpu-cores (approx 80% - 120% of 1000%).
The throughput or cpu-usage doesn't change when I use 2 or 4 or 10 or 16 R3load' processes in parallel. The "result" is always the same: approx. 3,5 GB Export-Dump and 1 CPU MAX-DB Kernelprocess is used !
so the BIG Question for me: WHERE is the bottleneck ?? (RAID-5 Disk-LUNS mirrored with LVM ???)
on HP-UX scsi_queue_depth_length I'v increased default value from 8 to 32 to 64 to 128 --> no impact
2nd question:
I'v read OS-Note 1327874 - FAQ: SAP MaxDB Read Ahead/Prefetch, and we are running MAXDB 7.7.07.19! (MaxDB 7.8 is not suppored via PAM for IA64 640-UC-Kernel) and as far as I understood, this parameter will no HELP me on EXPORT with primary-key-sequence ! is this correct? THUS: which parameter HELPS for speeding up Export-Runtime?
MAXCPU is set to 10, but ONLY 1 of them is used??
so this post is for ALL MAXDB GURU'S!
who will be intrested to contriubte on this "high-sophisticated" migration-project with 1.5TB MAXDB-Database-size and ONLY 24h Downtime !!
all tips and hints are welcome and I will give us coninued updates to this running project until WE did a successfull migration job.
PS: Import is not yet started, but should be done within vSphere 5 and SLES 11 SP1 on MAXDB 7.8 ....and of yours in parallel to export with migration monitor, BUT again a challenge: 200km distance from source to traget system !!!
NICE PROJECT;
best regards Alfred