TM1 9.4 MR1

Eric · Post by **Eric** » Wed Mar 11, 2009 4:39 pm

My IT group us pushing for the upgrade to 9.4 because of "SOX Compliance." Is 9.4 MR1 still to buggy to risk?

George Regateiro · Post by **George Regateiro** » Wed Mar 11, 2009 5:39 pm

We have been testing with 94 MR1 and have not come across anything too buggy that is going to stop our upgrade.

On a side note if their only reason for doing it is the SOX compliance then it seems like it is a weak argument. The new security groups are a start, but they are far from solving TM1s SOX compliance issues (at least as our auditors have defined our SOX requirements).

Eric · Post by **Eric** » Wed Mar 11, 2009 7:49 pm

I agree it is a weak argument, but that is the stance they are taking

Also I am not fighting them to much because I would like use a few of the new functionality 9.4 has.

Thanks for the comments about the bugs.

Alan Kirk · Post by **Alan Kirk** » Wed Mar 11, 2009 7:55 pm

Eric wrote:I agree it is a weak argument, but that is the stance they are taking Also I am not fighting them to much because I would like use a few of the new functionality 9.4 has.

Thanks for the comments about the bugs.

I note that you're still on 32 bit 9.0... are you going to be OK with the memory usage jump? (Which is admittedly greater in the jump from 9.0 to 9.1, but it ain't gonna get any better when you go to 9.4.) That's what's holding me off at the moment. We're fine on 8.2.12 most of the time... we have the occasional issue, but nothing much to speak of. But on 9.1+, then...

I definitely suggest that you check that out.

(Looking forward to your IT guys finding out that there are still no forced password changes in 9.4 or indeed any version aside from the kludge that has been posted in the Forum in the past... The SOX will definitely get lost in the washing.)

Steve Vincent · Post by **Steve Vincent** » Thu Mar 12, 2009 8:50 am

not to mention the vicious over-logged of metadata making auditing extremely difficult and log files the size of K2

Eric · Post by **Eric** » Mon Mar 16, 2009 2:48 am

How much of a memory increase are we talking for 9.0 to 9.1?

And then 9.1 to 9.4?

Alan Kirk · Post by **Alan Kirk** » Mon Mar 16, 2009 4:00 am

Eric wrote:How much of a memory increase are we talking for 9.0 to 9.1?

And then 9.1 to 9.4?

Ah, how long's a piece of string, compadre?

It's impossible to say because it's model dependent. This is from the 9.1 SP2 release notes:

Iboglix Release Notes wrote:In TM1 9.1, the TM1 Server delivers greatly improved concurrency and overall stability, as well as better diagnostics and monitoring capabilities. A side-effect of these necessary changes is that the TM1 Server may consume more memory than in previous versions. The amount of additional memory is primarily model-dependent. Customers who in previous versions have observed that they operate close to the memory limits of their hardware, should anticipate adding additional memory and/or modifying their models using documented TM1 memory optimization techniques. Customers who are currently using 32-bit TM1 Server near the 3 GB limit, may need to upgrade to a 64-bit TM1 Server. See the Applix Recommended Practices Repository on the Applix website (http://www.applix.com/rp/tm1_recommended_practices.htm) for articles on TM1 memory optimization and monitoring TM1 Server memory usage. Also, see the documentation on TM1 Counters in the TM1 Operations guide for more information.

Note the "may" consume more memory part. I think that this comes under the heading of "your guess is as good as ours, try it and see". We're running 9.1 on one box for Web only and 8.2.12 on the other; I've been planning to test what kind of difference is experienced with our model at startup on both systems, but haven't done so yet.

I believe that the step up to 9.4 involves a lesser increment. While 9.4 moves to Unicode, that's not really an issue for numeric data though it may affect some people who have lots and lots of strings stored but really... why would you?

Alan Kirk · Post by **Alan Kirk** » Mon Mar 16, 2009 4:59 am

Alan Kirk wrote:Note the "may" consume more memory part. I think that this comes under the heading of "your guess is as good as ours, try it and see". We're running 9.1 on one box for Web only and 8.2.12 on the other; I've been planning to test what kind of difference is experienced with our model at startup on both systems, but haven't done so yet.

I should have added; one of the reasons that I haven't made this a priority is that I'm not sure that startup memory usage is going to be a particularly valid comparison. If a lot of the change comes from changes to the locking model (note that one of the claims is "improved concurrency"), it may well be that you aren't going to see the real differences until you have users pounding the processors with read and write operations. That's what worries me more than anything; there's no realistic way of knowing exactly how bad the memory delta will be until you're actually staring it down in a production environment. You can guess, you can hope, you can extrapolate, you can simulate... but unless you're a long way from using all of your memory in an existing 32 bit environment, I'd be wary about 9.1+.

belair22 · Post by **belair22** » Mon Mar 16, 2009 6:25 am

Just to throw in my 2c....

Recently completed a fairly large corporate upgrade (both Server and silent client installs) - TM1 9.0 to 9.4 MR1 .

A fairly hefty TM1 model - Memory usage went up by only 10% - 15%. We have MaximumCubeLoadThreads set to 0 to conserve as much memory as possible.

Alan Kirk · Post by **Alan Kirk** » Mon Mar 16, 2009 7:29 am

belair22 wrote:Just to throw in my 2c....

Recently completed a fairly large corporate upgrade (both Server and silent client installs) - TM1 9.0 to 9.4 MR1 .

A fairly hefty TM1 model - Memory usage went up by only 10% - 15%. We have MaximumCubeLoadThreads set to 0 to conserve as much memory as possible.

Most interesting and useful information. Were you measuring that at startup, or as an average during the session's running time?

Either way, I think it may be enough to give all but the smallest 32 bit sites pause for thought...

mattgoff · Post by **mattgoff** » Mon Mar 16, 2009 11:33 am

belair22 wrote:We have MaximumCubeLoadThreads set to 0 to conserve as much memory as possible.

Didn't someone report that, although MaximumCubeLoadThreads does consume more memory during load, it frees it once the load is complete? If you're 32-bit and near the limit I could see being concerned, but if not, RAM is cheap (and freed RAM is free

).

Matt

Ken Vuong · Post by **Ken Vuong** » Mon Mar 16, 2009 2:25 pm

I have been testing 9.4 MR1 with the MaximumCubeLoadThreads and the RAM does not free after loading.
I could be mistaken, but I don't think the RAM can be freed.

Cheers,
Ken

lotsaram · Post by **lotsaram** » Mon Mar 16, 2009 10:17 pm

TM1 never releases memory back to the OS but it does internally recycle it's own pool of "garbage memory"

belair22 · Post by **belair22** » Tue Mar 17, 2009 6:05 am

I believe using the MaximumCubeLoadThreads to gain an increase in TM1 Service start time is a HUGE price to pay given the memory usage cost (which TM1 keeps and doesn't release back to the Server).

I've experienced models that utilise 65% of Server mem usage when not utilising MaximumCubeLoadThreads, the same model falls over 3/4 the way through a TM1 Restart where I've tweaked MaximumCubeLoadThreads to use 3 processors (# of processors -1).

Obviously if you have a shed-load of memory available you have no issue using MaximumCubeLoadThreads (for now).

TM1 Forum

TM1 9.4 MR1

TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1

Re: TM1 9.4 MR1