Server Load

[jorelb]

Our TM1 model is about 3 Gigs running on an x64 box with 36 Gigs of RAM and taking approximately 1 hour and 15 minutes to load. As we start to forecast for the remaining months (October, November, and December) and depending on the number of versions we end up for each of those months, I expect the server load time to increase significantly. I already experimented with MaximumCubeLoadThread but the memory usage trade off for the gain in load time is unacceptable. I also, on a regular basis, try to stream line rules and change rule calculations to TI processes. From a rule optimization perspective, I think I have done what I can and using MaximumCubeLoadThread will not work unless I can bump up our physical server memory.

However, I believe I saw an article in the Forum (which I can’t seem to find) of a technique that disables cube rules as the restart begins to speed up load time and enable the rules as the load is complete. I have identified the cube (and rules) that takes forever to load and would want to see if the technique will speed up our server restart.

Can anybody point me to the right post or assist in this query?

Thank you,

[LoadzaGrunt]

I believe I saw an article in the Forum (which I can’t seem to find) of a technique that disables cube rules as the restart begins to speed up load time and enable the rules as the load is complete

Have you looked at the DEMANDLOAD measure in the }CubeProperties cube. From the manual (Ops guide):

Indicates if a cube is automatically loaded when the server starts or is loaded 'on demand' only when a cube value is requested.

To 'enable the rules' later, you could look at the use of the ViewConstruct function in a TI process to schedule the request for a time outside of the server load window.

[tomok]

If you've got a 3GB model that is taking that long to load I would say you've more than likely got a serious overfeeding issue going on. Check out your feeders and make sure you 1) aren't feeding any cells other than rule calculated ones, 2) do not feed any cells that aren't really going to contain data, and 3) DO NOT FEED WITH CALCULATED ELEMENTS AS YOUR SOURCE!!!!.

[Alan Kirk]

If you've got a 3GB model that is taking that long to load I would say you've more than likely got a serious overfeeding issue going on. Check out your feeders and make sure you 1) aren't feeding any cells other than rule calculated ones, 2) do not feed any cells that aren't really going to contain data, and 3) DO NOT FEED WITH CALCULATED ELEMENTS AS YOUR SOURCE!!!!.

I think on point 3 you mean "do not feed with calculated elements as your target". Feeding from a consolidation shouldn't make a great deal of difference and is often necessary, especially for Detail -> Summary cubes. Feeding to a consolidation means that you're effectively feeding every element below it.

[tomok]

I think on point 3 you mean "do not feed with calculated elements as your target". Feeding from a consolidation shouldn't make a great deal of difference and is often necessary, especially for Detail -> Summary cubes. Feeding to a consolidation means that you're effectively feeding every element below it.

No, I meant what I said. If you feed using a consolidated element as the source then TM1 has to calculate that element in order to process that feeder. This can lead to some seriously slow load times because you are killing the sparcity of your model by effectively "building out" all the consolidated nodes and caching them in RAM in order to do the feeder. I agree that sometimes it is unavoidable but IMHO it should be avoided whenever possible, especially if server load time is an issue.

[lotsaram]

I think on point 3 you mean "do not feed with calculated elements as your target". Feeding from a consolidation shouldn't make a great deal of difference and is often necessary, especially for Detail -> Summary cubes. Feeding to a consolidation means that you're effectively feeding every element below it.

No, I meant what I said. If you feed using a consolidated element as the source then TM1 has to calculate that element in order to process that feeder. This can lead to some seriously slow load times because you are killing the sparcity of your model by effectively "building out" all the consolidated nodes and caching them in RAM in order to do the feeder. I agree that sometimes it is unavoidable but IMHO it should be avoided whenever possible, especially if server load time is an issue.

Actually that's not correct. Feeders are implicitly leaf level only. Feeding from a consolidated element is simply shorthand for feeding from an array of all N level descendants of that consolidation. The rules engine does not calculate the consolidated node in order to process the feeder. The area statement for a feeder is therefore a little different from a rule. If feeding from a high level consolidation in a large dimension the blow out in rule save and server load time is not from the consolidated value being calculated but from the number of leaf level cells being evaluated.