Page 1 of 1
TI for clearing zeroes in cells
Posted: Mon Aug 02, 2010 3:45 pm
by Ajay
Hi All
I'm looking to see if anyone could recommend some TI which runs nightly and simply removes cells with an inputted zero.
I'm investigating whether blank cells (ie those never been used) and zero cells (those where a number may have been sent) have an impact on the performance of the cube.
I'm about to restructure part of a cube and understand whether there maybe an impact would be useful before i do a single thing
Thanks
Ajay
Re: TI for clearing zeroes in cells
Posted: Mon Aug 02, 2010 5:31 pm
by Wim Gielis
Ajay
Both empty cells and cells with a 0 are the same for TM1. Moreover, they do not exist as such.
If you right-click the cube, you can unload it from memory. Maybe that can be interesting at times.
I really do not see what you mean, frankly speaking.
Wigi
Re: TI for clearing zeroes in cells
Posted: Mon Aug 02, 2010 8:52 pm
by Alan Kirk
Wim Gielis wrote:Both empty cells and cells with a 0 are the same for TM1.
Not entirely; if that were the case, the Undefvals rules statement wouldn't work. However if Undefvals isn't used, I'm not sure whether the server keeps track of the difference. (Indeed you're probably right that it doesn't.) I suspect that Undefvals may have been what Ajay was thinking of when he posted the question.
Unless this has been changed recently the server won't release the memory even after the values have been zeroed out but will keep it in "garbage" to be reused later. Wim's suggestion of unloading the cube will set it back to its original state when it's reloaded (assuming that it
can be unloaded; some rules and feeders will prevent that which is also why such cubes can't be "load on demand"), though I doubt that any performance gain would come from that. It'd certainly be advisable to reload it before anyone requests data from it to prevent them from experiencing the time lag while the cube reloads. More importantly, all cached calculated values would be dumped, which would also be a problem if the server itself was restarted.
In short, I think that Wim's right; this avenue is most likely to be a "dead end" performance-wise.