PA Cloud Optimization

[jim wood]

Guys,

We are running PA 2.0.7 and CA 11.1.4. We're seeing really slow performance when running a report / view containing one of our larger dimensions. The dimension has roughly 4000 members. In the cube viewer when opening the subset in TM1 Web it takes a couple of minutes to come back. When opening a prompt page in CA takes the same, some times more.

Our TM1 Server is local, both CA (CAM) and TM1 Web are on the Azure cloud. Our cloud team are looking at network optimizations to try and speed things up. But I was wondering if there is a setting to change that can optimize this? Currently we don't have a license for Workspace so that's not an option right now so we can't use that. I know Web is going away but I'd be surprised if I'm the first to hit something like this,

Jim.

PS. We've asked IBM support (chocolate fire guard comes to mind) and searched the IBM website without turning up much. It might be there but we've not been able to find it.

[paulsimon]

Hi Jim

Hope all is well with you.

We are running PA 2.0.5 so even older and we have dimensions with 13000 elements and we are not getting a problem in TM1 Web.

Is there anything special about this dimension? Some things that you might want to check :

Does the dimension have chained consolidations such as those you might use to do cumulative to date. PA doesn't cope will with this due the MUN issue.

Are there large numbers of subsets on the dimension such as can be generated if someone created subsets with a timestamp each time and never tidied up?

Are there large numbers of attributes on the dimension? Do any of those attributes have complex rule calculations?

Do you get the same issue if you open the dimension in Perspectives?

Regards

Paul Simon

[jim wood]

Hey Paul,

Thanks for getting back to me so quickly. All is well at our end just a little bit snowy today. We've about 1 1/2 foot of the stuff today so far and it ain't going to stop snowing until tomorrow. So I may be living in a igloo by this time tomorrow

I hope all is well at your end?

As for the dimension let's have a look see....

Does the dimension have chained consolidations such as those you might use to do cumulative to date. PA doesn't cope will with this due the MUN issue.

The dimension is pretty simple hierarchy wise. It has three levels in the only hierarchy: Project, Facility, Department & Total. It's not ragged in anyway and there is one to one relationship within the whole hierarchy.

Are there large numbers of subsets on the dimension such as can be generated if someone created subsets with a timestamp each time and never tidied up?

I can't tell if there are many private subsets but we currently only have 21 public subsets. Of those 17 are MDX based and one currently is empty. (I mentioned the last bit as we think it may have caused issues with the server over the last couple of weeks)

Are there large numbers of attributes on the dimension? Do any of those attributes have complex rule calculations?

Currently there are 40 attributes for the dimension. Of those 29 are rule based. The rules are not massively complicated but they do have at least one if statement in them. We have a project attributes cube and most of them are pulled from there and they include a filter to only pull values for the current year.

Do you get the same issue if you open the dimension in Perspectives?

No we don't. We didn't get the same issues in 10.2.2 before the upgrade, but then again the old install was completely local so I don't think we can blame that too much,

Jim.

[paulsimon]

Hi Jim

It doesn't sound like there is anything unusual in the model. The one thing you could try is perhaps temporarily knocking out the effect of any rules on the ElementAttributes and the Picklist cube by using an [] = STET. That would at least allow you to eliminate that as a possible cause.

I guess you need to look at the network side. We aren't using Azure but our servers are in the west of England and are accessed from London and various other places around the country, so the network latency and the need to go through firewalls is similar.

Perhaps the one difference is that both our TM1 Server and the TM1 Web Server are in the same location, whereas in your scenario you have two lots of latency both from the client to TM1 Web on Azure and then from TM1 Web to get back to your Local Server.

Are you using CAM? If so, where is your CAM Cognos Server?

Ever since upgrading we have had double login issue when using the TM1 Web URL API to launch a sheet. That can add 5 seconds or so to the sheet launching.

Do you have a problem with the View/Sheet in TM1 Web or is it only with the Subset Editor?

Regards

Paul Simon

[jim wood]

Hey Paul,

We are using CAM and it is located on the same server as both CA and TM1 Web. We have noticed some slow down with templates but that has been less consistent than the subset, so we've put that down to the pipe width which we have just increased. I will look to do the test with STET. The problem is we can only do it out of hours when the cloud pipe is pretty much unused,

Jim.

[Bakkone]

Hi,

Try switching to native logon. If that works better you know its a CAM issue.

More questions:

Is it a windows server or linux?

Are you using IIS?

Are you using SSL? If so for which traffic. Maybe try turning it off for the CAM authentication to see if it helps.

[jim wood]

Try switching to native logon. If that works better you know its a CAM issue.

I will on a non prod server and see if it makes a difference.

Is it a windows server or linux?

It's windows for all of them.

Are you using IIS?

Yes IIS is being used for both CA and CAM

Are you using SSL? If so for which traffic. Maybe try turning it off for the CAM authentication to see if it helps.

No we're not currently using SSL, but we are looking at adding it soon.

Thanks for getting back to me,

Jim.

[ykud]

Our TM1 Server is local, both CA (CAM) and TM1 Web are on the Azure cloud. Our cloud team are looking at network optimizations to try and speed things up.

I think TM1web still uses the chatty old protocol with multiple trips to get data from Tm1 server, so you'd see this when you're increasing the latency by moving servers further away. I'd say you want them back close together (either move Tm1web back local or move tm1 server to Azure).

CA performance should depend on whether you use TM1 data source type (old API) or Planning Analytics (REST) -- maybe try switching the datasource type to see whether it'll fix the prompt performance (it should)? There's a few changes you need to do for using Planning Analytics datasource (main one being aliases), but it's a fairly straightforward switch.

Cheers,
Y

[jim wood]

Many thanks. We're looking to move the TM1 services to the cloud over the next couple of weeks. I'll also ask our CA guys to have a look at the data sources. I think we're on PA but I'm not 100% certain,

Jim.