10.1 - Persistent Feeders and PI clash at startup?

[garry cook]

I've been seeing some pretty serious increases in load time on a model (48 mins to over 6 hours) and having narrowed down the potential issues by running the same base data files / config files on different machines, it appears that I've narrowed the issue down but not sure if others have experienced this or if it's peculiar to us.

Operating on 10.1 (non SP'd) with conditional feeding taking place in a few areas (trying to get rid) so only using 1 core on startup.

I can pretty consistently make the following behaviour happen with the only difference being the three variables below -

PI (Parallel Interaction) = F, PF (Persistent Feeders) = F, no .feeders files = 48 mins startup
PI = F, PF = T, no .feeders files = 48 mins
PI = T, PF = F, no .feeders files = 48 mins (seeing a pattern here?)
PI = T, PF = T, .feeders files exist = 7 mins
PI = T, PF = T, .feeders files do not exist = 6.5 hours and counting (98% of startup RAM completed).

It's getting jammed up on the cubes where there are conditional feeders and the area it's taking a huge amount of time on is processing the feeders for the associated rule files. Not sure if the conditional feeders is a co-incidence as they're also some of the larger cubes so may be a red herring.

Where PI = T, PF = F, I have as an example one rule file which is taking 98 seconds to generate feeders and cube dependencies and same model, same config except where PF = T with no feeders file, it's taken over 18 minutes to do the same task.

Has anyone else seen this behaviour?

[jim wood]

I've said this on a few threads now. 10.1 is riddled with bugs. You really need to move to 10.1.1 FP1. All of our customers on 10 have moved to this version as it is the most stable version of 10,

Jim.

[garry cook]

Yep, moving up to .1 SP1 is on the radar (although i'm far more keen on 10.2 when it finally comes out of beta) but this is the worst bug I've encountered by far so just wanted to check if it was us or consistent to the version.

[jim wood]

Do IBM have a log of the raised issues? I know that we encountered a lot of issues at J&J with 10.1 and they were raised with IBM. I would check it out with them to see if this issue has been raised and to see if a fix is in place within 10.1.1 FP1.

[garry cook]

If they have, it's proving difficult to find

Will be raising one directly but would be interested if this behaviour has been seen in 10.1.1 by anyone else.

[stephen waters]

would be interested if this behaviour has been seen in 10.1.1 by anyone else.

Garry,
David has a keen interest in this area and may well have some sort of answer. He is on his way back to Uk after an extended Vision 2013 visit and should be able to respond in next couple of days and I will send him link in case he misses it.

btw, following Vision conference we are really looking forward to what may be included in what could be called TM1 10.x that is possibly due to be released sometime in the indefinite future. (sorry, NDAs prevent me from being too explicit!). Seriously, there looks like some really good stuff both front and back end, should be worth waiting for....

[garry cook]

Thanks Stephen, much appreciated as getting some server space to run this model is at a premium to do a 10.1.1 test. I've not got as much faith in Jim in IBM's response about what's fixed in various FP's having had my fingers burned in the past (no offense Jim)

In terms of 10.x and the changes, I'm in full agreement assuming some of the key things that they may or may not be talking about in terms of potential things that may or may not be happening to make more or less usage of insert 3 letter acronym here for hedgehogs and their like do actually work. I was sceptical about PI and proven wrong so I'm prepared to be blown away by the awesomeness that is 10.x and it's multiple sproglets.

In the meantime until they do release 10.x though (or more likely 10.x.2 SP3), I can live with most of what we've found in 10.1 but if this PF / PI clash is fixed in 10.1.1 FP1 then I guess it's another server side upgrade on the cards in the meantime as this is a pretty shocking omission from testing

[jim wood]

Garry,

My optimism isn't based on what IBM have told us but based on the experience of one of our clients. They had 10.1 and moved to 10.1.1 FP1 and the large majority of the issues have gone. They recording much higher stability rates. I can pass you on to one of our guys that can give you details if you're not convinced,

Jim.

[whitej_d]

A similar sounding issue relating to PI was fixed in HF4 for 10.1, although poorly documented. We were finding that freeze processes were taking a long time with attribute driven feeders. Installing the HF fixed it.