CalcCache settings

Question

Hi Experts,
I have recently set up a new system 9 environment (9.3.1). I currently have 2 servers 1 blade 64bit with 1tb of space and 32gb ram this is for our production essbase server, and 1 32bit with 500gb disk and 8gb of ram for our shared services environment. This is a massive improvement from our old 7.1.5 system that only had 4gb of ram on a 32bit platform. However as at the time we were limited as to what we could do calculation wise on the old server, we no longer have these restrictions.
I have migrated all our application to system 9 and found that retrievals arnt much better? I checked out the essbase config file from our old 7.1.5 and its been set as follows:
; The following entry specifies the full path to JVM.DLL.
;Original JVM entry: JvmModuleLocation d:\Hyperion\Essbase\java\jre\bin\hotspot\jvm.dll
JvmModuleLocation D:\Hyperion\common\JRE\Sun\1.4.2\bin\client\jvm.dll
;Override default Direct I/O to specifically set Buffered I/O access
__SM__BUFFERED_IO TRUE
__SM__WAITED_IO TRUE
EssbaseLicenseServer @UKBSS016
AnalyticServerId 1
;Set data block locking defaults
CALCLOCKBLOCKHIGH 200000
CALCLOCKBLOCKDEFAULT 100000
CALCLOCKBLOCKLOW 5000
SET LOCKBLOCK HIGH;
;Enable the calc cache and set defaults for memory use
CALCCACHE TRUE
CALCCACHELOW 5120
CALCCACHEDEFAULT 10240
CALCCACHEHIGH 20480
CALCNOTICEHIGH 50
CALCNOTICEDEFAULT 20
CALCNOTICELOW 5
; Set Dynamic Calculator Cache
DYNCALCCACHEMAXSIZE 512M
DYNCALCCACHEWAITFORBLK FALSE
;DYNCALCCACHEBLKTIMEOUT 0.005
;DYNCALCCACHEBLKRELEASE TRUE
;DYNCALCCACHECOMPRBLKBUFSIZE 12540000
;Set maximum number of data load errors recorded to error file before logging stops
DATAERRORLIMIT 1000000
I have tried to apply these settings to my new server as the config file was empty with the exception of the jave module and what not. No specific performance settings.
I was wondering what sort of scop I have to increase these settings, and what would happen if I set any of the above options too high?
A lot of the issues arise from formula in my outlines attached to dense dimensions but have cross dimensional operators to sparse dimensions. so quite intensive, but I have no choice in changing this design of the outline with the exception of making a few of the dense upper levels dynamic calc. but want to be able to get the most out of the new machine aimed at retrieval.
Any suggestions/help would be great. If you need any more information them please ask
Thanks in advance
Eddie

Answer 1

Hi, yes that would be great......
I am basically running a calc all on 4 of our databases that contain a total of 5 dimensions listed below, These data bases are then transparently partitioned to a larger database that contain additional functions, such as working out MAT etc.
Year DB's
Dimension type members in dimension members stored
Period - Dense - 836 - 775
Fact - Dense - 152 - 118
Promotion - Sparse - 48 - 47
Brand - Sparse - 2801 - 2038
Customer -Sparse - 5314 - 4845
the above are the yearly databases dimension structure
Cache sizes are as follows:
Index 16384kb
data file cache 32768kb
data cache 614400kb
Statistics
number of existing blocks: 141202
Block Size 731600B
Create blocks on equation is set on.
Transparent Partitions DB
Period - Dense - 836 - 775
Fact - Dense - 152 - 118
Currency - Sparse - 5 - 4
Year - Sparse - 6 - 4
Accumulation - 18 - 1
Promotion - Sparse - 48 - 47
Version - 57 - 30
Brand - Sparse - 2801 - 2038
Customer -Sparse - 5314 - 4845
Index 153600kb
data file cache 8192kb
data cache 307200kb
Statistics
number of existing blocks: 0
Block Size 731600B
Create blocks on equation is set off.
In the transparent model formula exists to work out MAT's
any help would be greatly appreciated