TM1 Forum

Hi guys, long time lurker, infrequent poster.

We've just had two of our server boxes upgraded in anticipation of moving from 9.5.1 32bit to something a bit (more) modern. What's puzzling me at the moment is that one box takes over twice as long to start up as the other (48 vs 21 minutes) -- and the TM1 configuration is exactly the same (they are mirrors of each other). We don't have a lot of access to the physical servers but when I started to investigate I found that the slower server actually has more cores (24 vs 16) and more RAM (256 GB vs 32 GB). I can't see any different software installed on either box and they're both on the same version of Windows 2008R2. We actually have another virtual server in another location that's also starting up in around 21 minutes too, which is closer in spec to our "fast" server.

Now I don't think the cores will make a lot of difference given we're still on 32bit TM1 but I'm having a hard time understanding the speed difference on load. Once up and running they seem fine, no noticeable difference between them.

The plan is to upgrade to 10.2.2 64bit (service pack infinity) which with feeder files and multi-threading will give us a big boost but if there's a fundamental flaw in the configuration of the system I'd rather explore now before we upgrade. I asked our IT people why the more powerful system was slower and they came back with the helpful answer that one system had more cores and RAM than the other but literally just listed the specs so I'm none the wiser.

More cores may not mean faster! If TM1 is starting using a single thread, the newer box may be using one core that happens to run slower than the older box.
Could also be anything from paging files to an over-zealous anti-virus scanning every log entry.

Would be useful to see what is in your TM1S.CFG to give some better insight otherwise it is just wild guessing.

Hi
I agree, it would be useful to see the cfg file, and it would also be good if you could take a look on the tm1server.log file. Are all entries slower, or just specific feeder calculations?

Sorry for the late reply - I only get to look at this forum once a week.

The two cfg files are identical (barring server name differences obviously) as they are literally copies of each other. We're on 9.5.1 so I don't think the parameters around multiple threads / persistent feeders or any of that good stuff are applicable.

Good idea on comparing the two log files, will start up staging (it's normally dormant just waiting to be started up as a back up) and then compare the two timings. If it's a consistent timing difference on the feeder processing steps it would point to the core being slower.

Thanks

If your servers have a different number of cores, are you sure, that you run the same processor architecture with same clocking?
This is the first thing to make sure, as both criteria, cube load and feeder processing, have a direct link to these properties.

Hi, update for you all, thanks for looking.

We have now established we do indeed have a production server with slower cores than our staging server. We did some calculations based on IBM's PVUs and the timing differences in the log files so that all seems to stack up. My original concern was that there was a server configuration issue (in the OS) and that we wouldn't notice it until it was too late on a newer version.

Theoretically this new system will give slower response time to an individual user (as each gets a thread) but I doubt it would be measurable / noticeable (except for admin type rule re-compiling etc) so I'm not so worried about that. My next question to our IT folk is how exactly do they think a staging machine with 32GB of RAM is supposed to act as a "backup" for a production machine with 256GB of RAM. Of course if I push too hard they'll just take the extra RAM off production so I'll step carefully on that one (where's the headbanging emoticon?).

Tony