Write speed during Local Checkpoints

[Jörg Nowak]

Hello everybody,

we try to speed up the start up process of a big cluster (5 millions of user profiles in 6 tables -->30 million records). The phase which writes the local checkpoint during start up consumes ~50 % of the start procedure. (13 of 23 minutes).
To increase the write speed we use a Storage Area Network (SAN) an we configured RAID 10.
During tests (with checkpoint during restore) we get ~13 MB write speed per sec.
But during the local checkpoint during normal start up we reached only ~ 4 MB per sec.

How can we speed up the local checkpoint during normal start up.

Any Ideas ?

Regards
Joerg Nowak



[COMPUTER DEFAULT]

[DB DEFAULT]
#ANY CHANGES IN THIS SECTION WOULD HAVE TO RESTART THE NDBD DAEMONS. ONLY THEN THE CHANGES WILL HAVE EFFECT.
NoOfReplicas: 2
DataMemory: 4G
IndexMemory: 500M
Diskless: 0
#0 means Data is written to the disk. 1 means that data is handled in memory itself.
StopOnError: 0
MaxNoOfConcurrentOperations: 260000
TransactionBufferMemory: 5000

#Disk Write parameters
NoOfDiskPagesToDiskDuringRestartTUP: 100000
NoOfDiskPagesToDiskDuringRestartACC: 100000
NoOfDiskPagesToDiskAfterRestartTUP: 1000
NoOfDiskPagesToDiskAfterRestartACC: 1000
BackupDataDir: /locA/db/backup

#Performance parameters
#TimeBetweenLocalCheckpoints: 16 #Meaning: Checkpoint runs continuously
NoOfFragmentLogFiles: 300 #Meaning: 4096MB of REDO Log Space
RedoBuffer: 2048M
# MaxNoOfAttributes - Number of columns over all tables
MaxNoOfAttributes: 2000

[API DEFAULT]

[MGM DEFAULT]
ArbitrationRank: 2

[TCP DEFAULT]
PortNumber: 58004
SendBufferMemory: 10240000
ReceiveBufferMemory: 10240000
..........................................