9.1 SP3 Instability Issue

[David Usherwood]

Interesting. However:

a In your (production) environment, does the new locking approach _actually_ run better than in earlier versions?
b I built a set of testbeds to test _input_ performance and ran them against various versions. 9.1 didn't do well at all. 9.4 did rather better, but not as well as 9.0. I have passed the testbed and my findings to Cognos (and also round others in the TM1 community - to validate the approach and the code). This was a bit of a laboratory approach, but it did test multi-user input using VBA running on multiple PCs with some randomising of timings.
c Cognos recommended splitting input into multiple microcubes (which I folded into my testbed). Are you using that approach?

[jim wood]

Multiple micro cubes eh? Unfortunately not. The data being loaded is budget version data from EP. The EP extracts are rather large. I am going to be contacting Cognos to find out if there is better way of getting EP in to TM1 than text export. Also this budget data needs to be in one cube for review. I'm not sure how micro cubes would help in this situation,

Jim.

[Eric]

I'm on 9.0 SP2 to I do not use Action buttons yet, but the need is rising.

Regarding locking issues. We have users input data in excel, then they execute some VBA that appends an ascii file which TI Uploads hourly or via a power user manually executing the TI. This eliminates most locking issues.

[David Usherwood]

Aha!....
Jim, if you are loading big flat files, indeed, micro cubes are irrelevant. (I didn't like the idea anyway, but Dave Corbett said that's what the new locking model needed to work best.)

But I did have a not dissimilar issue recently, with a client with a pretty big system. They wanted to create a static slice of their calculated budget ie to copy the ruled values into an unruled slice of the same cube. We built a copy routine and it took....
3 days
to run. Not really acceptable.
I found a tip on the old best practices forum, tried it out, and it worked very well - time came down to 12 minutes.

a Export out the values you need to import (in your case, that's done)
b Use RuleLoadFromFile in TI to load a _blank_ rules file against the cube
c Import the flat file
d Use RuleLoadFromFile to restore the rule.

I know this isn't quite what you are trying to do, but I suspect that your destination cube is ruled. Try creating a rule-free clone and loading the file in there. I would assert it will load far quicker. On my Dell D820 I get 40,000 records per second. I've had to take PoC files supplied by prospects and artificially blow them up so I can show them that something's actually happened.

I may not have a full idea of your setup - do feel free to respond. Also, your setup is quite relevant to the new, all-Cognos world (yeah...) so it is worth getting the issues nailed down.

Look forward to your response.

[jim wood]

Hi David,

You may be on to something. I have checked this weekends log file and the most amount of time is taken loading data from my sandbox cube in to the main data cube. The data comnig out of the sandbox is huge. It has no rules but lots of data. As I mentioned it has no rules but it still takles up to 9GB of RAM. The process I currently have uses a view from the sandbox. May be I should split this in to a file export then load in to the next cube.

The process you have mentioned, does it have any impact on performance?

Jim.

[David Usherwood]

The approach I described (not my invention, as I said - from the old best practises area) did have a major effect on performance (3 days down to 12 minutes). You describe a somewhat different approach ie export to file and read back (somewhere else). I think that that can help - but does your destination have lots of rules (specifically feeders)? That's where the optimisation will be most seen.

[jim wood]

Hi David,

The target does have a few rules at the moment. I am however going to be doing something about that soon. I'll see if this speeds things up first,

Jim.

[jim wood]

Guys,

We have finally had something back from Cognos. They have told us that there is a compiler issue with the Solaris base issue of 9.1 SP3. They have asked me to download and install 9.1 SP3 U2. They have assured me that this version solves the issue. Let's hope so,

Jim.

[David Usherwood]

Result! Well done.
Still think 9SP3U4 or 7 would be fine.

[jim wood]

Result! Well done.
Still think 9SP3U4 or 7 would be fine.

We'l give this new version a go. If it is no more stable we may have to revert back to an earlier version. We are however seeing this as a last resort as we then have roll back the clients. This may not seems like a big job but with my company getting IT to flush the toilet is a result!

Jim.

[jim wood]

Another update from Cognos. It seems that 9.1 SP3 U2 has a unrelated complier error in it. They have now told me to download 9.1 SP3 U1. Looks like we are going to be playing minor version bingo.

[Martin Erlmoser]

does anybody know anything about readersbypass writers with the new locking model? +9.1