And as of 1st of November I’ll be joining small, but proud ”bunch of guys that share the same passion and dedication for Cognos (and sometimes the same sick humour :))” ©, namely, PMSquare.

There’ll be more upside-down stories here, I promise.

As they say it here, “Cheers to you all”

Reasons vary, but mostly it’s that I’ve been, you know, busy-busy. As I now start to reflect on it, it’s always a point of view thing and a question of self-control and ability to say no )

But enough philosphy, brief recap on what happend while I was out “there” in real world. I’ll divide this into two parts (by vendors ))

Big Blue

Made some POCs in autumn, nothing that special (some neat integration, but not cognos-related). Carried on a couple of projects with  EP+BI typicalities. All in all, nothing that new.

BUT I’ve finally completed the other certification path there is at Cognos\IBM, notably, the “BI Professional” road. It’s a deal longer, as of now, you have to pass 3 “role-based” exams, 2 mandatory (Report Author and Metadata Model Developer) and one of you choice to get a chance of passing BI Pro exam. It’s used to coast 10k, which could be cut down to reasonable 200$ if you attended a Cognos Workshop (which costed 2,5k), but thanks to IBM that’s gone and now it’s 200$ from the start. Just opened certifications link and I’m, frankly, quite surprised with new IBM names for old Cognos certifications. According to this I’m:

• IBM Certified Designer — Cognos 8 BI Reports
• IBM Certified Developer — Cognos 8 BI Metadata Models
• IBM Certified Developer — Cognos 8 BI OLAP Models
• IBM Certified Solution Expert — Cognos 8 BI

and it turns out I was an

• IBM Certified Solution Expert — Cognos 8 Enterprise Planning

instead of Cognos EP Professional.

Wording is something you cannot easily take from IBM ) Will put that list in about page, just to get more mail asking for exam questions.

Hint: In comparison with the old Modeler exam, current one is easier.

And in the first 2 months of this year we did a “nightmare-style” POC with EP + TM1 + BI, where I could finally get an idea of what TM1 really is and fully realize some ideas like incremental administration links and, therefore, “real-time” EP -> TM1  data propagation. I’m planning to write about this in separate posts, though. And, boy, it was hard, and we flanked the final presentation due to me being ill and bla-bla, and I’m still recovering breath from this.

But enough about IBM, let’s turn to

Oracle

We’ve made a really nice billion-facts cube with our colleagues from “SportMaster” and talked about it in exceeding detail on Russian Oracle Technical Forum 2008 and Oracle BI Forum 2009. Essbase turned out to be a good engine for this kind of task and the moment when it finally kicked-in on 100gb dataset will retain as one of best in 2008. The fact that one of dimensions was a million rows parent-child didn’t raise that much of a problem.

Moreover on another site we duly built dimension outline with around 30 mln members. Although duly is not the right word, “kicking and screaming” would be a better wording ) And the whole area of trying to tune something in underlying Berkeley DB is still open. But I must admit, an 8 gb outline file is really impressive.

So I’ve spent pretty much time toying around with essbase (except when I wasn’t with TM1)). And I’m also a Hyperion Essbase 9.3 Developer Certified Expert (I won’t enter Prometric for a while, I hope).

So it was a rather crowded time and I hope that nothing will stop me from blogging some details and future encounters. At least, I’d try my best to. Cheers to your patience, folks )

Anyway. I’ll be spending next couple of weeks in Europe, so if you want to share a beer — drop a line.
I’ll be starting 29.07 in Amsterdam with no set programm afterwards (although Berlin seems a definite place-to-go).

1 This blog changed hosting once more. Now it’s ultra speed webfaction instead of Yahoo.
Speed, ssh, svn are main switch reasons

2 I’ve started a more light-weighted blog in russian ykud.ru (who’d guess).

3 And I’m thinking about this blog’s future. I’ve got only one post laying in drafts (it’s called “Testing in EP”) and after publishing it I’ll be out of ideas to write about. Temporatily, I hope.

Maybe I’ll publish my current OLAP-research ideas, since I’ve already been beat-up by some china research group.

I really expect and encourage comments on this naming proposal. Maybe together we will create something like Sun Java Code Conventions )

Whoa, it’s a really huge document, maybe I’ll it finish someday…

[1]: http://www.amazon.com/OLAP-Solutions-Building-Multidimensional-Information/dp/0471400300

Task is quite simple&straightforward: calculate number of periods in which current stock will be sold in a store. Like: today we have sales enough for 3 weeks. That’s a very common sales demand analysis parameter (Months of Sales (MoS), Weeks of Sales (WoS), DoS and etc).

Formula is quite simple: Current Stock / Average Sales over some period

But there were some difficulties in my case.

I’ve got only partial data, some periods were completely missing, and the requirement was to treat missing sales data as zeros. And I also had to get the last known stock as current stock, because stock data could be missing too.
Last stock is pretty easy to get using last_value oracle function, but getting average sales is hard. You cannot use fixed window width in partition by, since data can be missing. Moreover period varied other the years, so I’ve ended up adding up a field in time dimension called periods_ago, that contained time_key reference from which to sum sales.

So I’ve started with simple subqueries

SELECT
        time_key,
        product,
        company,
        sales_qty,
        (SELECT SUM(sales_qty)
           FROM dwh.sales sales2
         WHERE
            sales2.product = sales.product,
            sales2.company = sales.company,
            sales2.time_key BETWEEN weeks.periods_ago AND sales.time_key
        ) AS running_sales_sum
FROM
        dwh.time_dim weeks, dwh.sales sales
        WHERE
           weeks.key = sales.time_key

and that is terribly slow on 10^6+ rows. ~20 minutes.

But with model by it’ll be like:

SELECT
        TIME,
        product,
        company,
        sales_qty,
        last_stock_qty
FROM
        dwh.time_dim weeks, dwh.sales sales
        WHERE
           weeks.key = sales.time_key
      model
        IGNORE NAV
          PARTITION BY (product,company)
            dimension BY (time_key)
            measures
            (
                periods_ago,
                sales_qty,
                0 eight_weeks_sales_sum_qty,
                NVL(LAST_VALUE(STOCK IGNORE NULLS)
                OVER (PARTITION BY COMPANY, PRODUCT, SEGMENT, DISTRIB_LEVEL, SYSTEM
                ORDER BY TIME ASC),0) AS last_stock_qty
            rules   UPDATE
            (
                running_sales_sum[ANY] = SUM(sales_qty)[TIME BETWEEN periods_ago[cv()] AND cv(time_key)]
                            END
			)
 
 
            )

around 3 minutes ) Notice the last stock calculation.

Here’s a couple links to good model by examples:
A wonderful tutorial at SQLSnippets
Official documentation restriction section — read that carefully

Some stats: around 10 unique hits a day to total of around 4k, quite as much feed readers, mostly from USA&Europe. And from Australia, Taiwan, Japan even New Zealand  and more-more-more. I expected more local readers, but it’s even better for me. Almost no direct competitors reading )

Since I’m mostly doing complete EP+BI projects now, MDM problems itch the most. So expect some post on this theme.
And some on huge elist problems (I’m testing a 15k elist app currently).