|
From: Heidi N. <HEI...@de...> - 2001-07-18 16:31:18
|
Hi ,
the goal of this mail is to start a discussion about how
to handle mass operations . It is not always the best
solution to avoid them .
At the moment I'm running in such a mass problem with
the instrumentation of "Filesystems & Volume Management"
and I'm not sure how to solve the forced crashes .
PROBLEM DESCRIPTION
- affected module : sblim-fsvol-0.6.1 ( tar.gz )
FSVOL_0_6_1 ( CVS tag )
- affected part : the instrumentation of the Logical
Volume Manager for Linux (LVM)
- affected class : LVM_PVExtent
Each instance of the class LVM_PVExtent represents one
physical extent in the Linux Logical Volume Manager (LVM) .
- provider code :
* com/ibm/wbem/linux/device/LVM_PVExtent.java :
calling the NPI - uninteresting in the problem
environment
* LVM_PVExtent.c :
1 step : enumeration of all physical volumes
-> 6 physical volumes
2 step : for each instance of the physical volume
all physical extents are collected in a list
-> ~ 4000 = sum of all physical extents
! this happens in functions of the tool
library CimFsVMgmt.c whithout any problems !
3 step : for each entry in the list of physical
extents one instance of LVM_PVExtent is
created ... no - should be created
This enumeration of all instances of the class LVM_PVExtent
brings on the problem how to handle this mass operation
and avoid crashes .
Now one discussion point can be why it seems to be
necessary to do such a "maybe useless" operation . It is,
from a systems management application view, really necessary
to get all the physical extents of one system to have a look
in the way they are connected to their corresponding logical
extents and the logical volume ? What might be a case where
this granularity becomes important ( Standard value fo a
physical extent is 4096 kB !!! ) .
Or shall there be other ways to offer detailed informations .
An other discussion point can be, how to solve such problems
whithout avoiding enumerations with a mass result . I' sure
that there are other environments where it is not possible
to strike out operations that have a mass result .
Now follow some possible solutions as discussion basics .
POSSIBLE SOLUTIONS
* increase the memory allocation size for the Java Virtual
Machine ???
* take this granularity out of the schema
( delete LVM_PVExtent ) ???
* more suggestions ???
Thanks for your attention ... Heidi
Mfg / Regards
Heidi Neumann
IBM Laboratory Boeblingen - Germany - Dept. 3243
LTC - - Linux Technology Center -- 7UFA
Systems Management
Phone : ++49 (0) 7031 / 16 - 3193
Schönaicher Str. 220, 71032 Böblingen
hei...@de...
--- try sametime to contact me ---
http://w3.opensource.ibm.com/projects/ibmcimom/
http://oss.software.ibm.com/developerworks/projects/sblim/
|