Urgent Help needed with Estimation

[bloggerman]

Hi,

I need to estimate a Cognos effort and need some help with the same.

Could you please help with some tips on how to go about estimating. If you have some template available that would be great.

Please feel free to send a mail at damnhellraisers@gmail.com

Thanks for the help. It is really appreciated

[UkCognosConsult]

Hi,

We really need a bit more detail, do you mean estimating man hours for report building, Frameworks and Data Manager creations and such?

UkCognosConsult

[blom0344]

bloggerman,

The forum allows for sending PM's for direct peer-to-peer contact. This may a better idea than posting your email addres openly within a post..

[bloggerman]

Yes, i have to estimate for a FM model and few reports out of it. I would need to provide and estimate on the whole as in from the point we start analysing to the point of deployment. I hoping to get some direction. 

Could you also suggest something for cubes.

[blom0344]

An FM model may be based on a single fact with a handful of dimensions, or it may cover an entire source system with hundreds of tables. A single report may involve just 0ne list or it may have 15 subreports within one report. You would need to give a bit more background details to paint the picture..

[bloggerman]

Here as below

3 Facts, 4 Dimensions

Report is 1 list. Drill through at 2 levels. You can drill through from one report to the second and from there to the third.
Each report has 7 odd columns. 2 from dims and rest 5 from facts.
One prompt
Minimal formatting.

I would think this would come under as Simple/Medium complexity?

-------------------------------------------------------------------------------------------------------

There is another set of reports i have.

One List, 8 Facts, about 7-8 dimensions.

Report has about 6 prompts, Has cascading prompts, Some Java Scripting
Drill Through from one report to the other. Good amount of formatting. Conditional&Render variables. 2 set of prompt pages each rendered based on a condition. About 15 columns.

Could you also guide me on the approach one should take for estimating..Some tips?

[blom0344]

I would build a test model in FM ,create the first report and then you have some reference material for future development. With no experience yet I would say reserve 3-4 days (including some testing) An experienced Cognos developer might spend 1-2 days on your first example.

Let me point out that these are quite 'humble' report designs..

As mentioned earlier , if this is all new stuff you should push for some time to 'play' around. do not aim for the final versions of either model/ report with the first attempt!!

[bloggerman]

Thanks.....

What would think a complex report would look like?

[bloggerman]

There are couple of other questions that  have jolted my confidence a bit. Since they are basic....

1) How to decide which to use. List vs Cross tab? Is it just a case of relational vs dimensional.

2) If i have a dimensional database should i always go for cubes? Or should that be based on only a need of OLAP style drill up drill down requirement from users?

[CognosPaul]

Lists and crosstabs are different beasts that can achieve different goals.

Since I primarily work with cubes, I tend to use crosstabs more. But crosstabs work just as well with relational data sources, just as you can use a list with a dimensional. The decision of which one to use should be based on the needs of the report.

If you need the data arranged in a tabular format, maybe as a source for an ETL process, a list is the way to go. If you need the data pivoted, such as years in the columns you need to use crosstabs.

Whenever asked if a project should use DMR or real cubes, I always say to go with cubes with a relational database behind it. Cubes are great for aggregating data, but tend to have poor performance when you want to return individual rows.

Lets say you want to find all of the customers who purchased more than $5k worth of glowstick chargers in July of 1983 and purchased less than $15K of of elbow grease in August of 1984. While you could do this in SQL, the query would be ugly and slow. On the other hand, that is exactly the kind of query that cubes are designed to work with. Now lets say that once you have that list, you want to see all of the data for a specific customer, all 300 fields arranged in a nice easy to read page. This would be impossible to do with cubes*. The SQL Query on the other hand would be nice and easy to read.



*Maybe not with the latest version of Analysis Services, people in the know are promising me the world.

[bloggerman]

Thanks Paul.

Also, last part my question was in relation to Relational model ( not DMR) vs Cubes/DMR. Infact once in an interview i was asked on how i would judge if a requirement should have normal relational design or OLAP style Cubes/DMR. My reply was it would depend on requirement of slicing/dicing, Drill up/drill down, need for summarized data vs detailed. The interviewer didn't look fully satisfied. Are there any other?

Also, Could you give me an example of a complex report and how much time you would think it would take to build

[CognosPaul]

My criteria for deciding relationalvs OLAP are simple. Will the report ever need to pull individual records? If so, then relational. Else dimensional all the way. There are simply too many things relational databases can't do, or can't do easily, to make me consider them for a range of reports.

The complexity of a report is very subjective. I have a report that has an output of a few hundred pages, but it's a relatively simple report (page set/crosstab). I have another report which involve several nested lists and pagesets that only works by referencing a parameter map inside FM.

The time it takes is also dependant on your requirements. Does it need to look good in Excel and PDF? Is it only going to be displayed in HTML?

The most complex report that I have was made using only repeaters, a crosstab and a few dozen HTML items. It took me close to two months to complete it, including the inevitable rejects and modification requests.

[blom0344]

I would consider another angle. Does the user need to answer an adhoc question that involves some complexity? Then a dimensional model will suit best. Almost any issue can be dealt with a relational model, but you do not want a casual user to write their own report.

Our largest management reports involve 20 report pages, some 50 queries with loads of conditional rendering to give the optimal PDF output. From highlevel to very lowlevel detail. In fact it is a totally fixed template for new data every period. Definitely something for a relational model and build by seasoned developers..