Thursday, August 20, 2015

The Plunge Into Hybrid Part 2

Yesterday I thought I had good results from my new Hybrid cube. I was running a MaxL script comparing ASO, BSO and Hybrid. Hybrid won in a landslide. What I neglected to do was connect via Smart View and do some ad-hoc analysis. I tried that today. The results... well, um, not so good.

My expectations of an Essbase cube are extremely high. If I'm not getting sub-second response on smallish queries I go back to the drawing board. My initial query on the Hybrid cube was 2.5 seconds. The same query in BSO registers 0 seconds in the log. While 2.5 seconds might not seem like a long time, it is to someone who expects instantaneous response. Also, it will spell disaster for larger queries. So it's back to the drawing board.

First I tried messing with some cache settings. That didn't help at all. Next I tried some unorthodox dense/sparse settings -- also didn't help. So what's next? There are a few possibilities:

  1. This is one of those *fringe* cases that requires ASO or pure BSO.
  2. I need to buy better hardware.
  3. We just need to accept slightly slower retrievals from Hybrid (at least for the time being).
  4. I'm missing some tuning trick.

The sum total of Hybrid documentation available from Oracle is about two pages in length. That's not much to go on. So what have people seen? Is Hybrid ready for prime time and if so, what are the tuning tricks?


Cameron Lackpour said...


I'm not sure if you meant the aggregate cache when you noted that you tested different cache settings but as of, the cache is 32 mb whether you like it or not.

Oh, you can use the MaxL to change it, but it doesn't take effect.

I've spoken to Steve about this and it's on their radar.


Cameron Lackpour

TimF said...

Hi Cameron,

I was testing out data cache settings and wasn't noticing any change in performance. I was wondering if they'd have any impact on the speed of aggregations in the sparse dims (i.e. having more blocks available in memory). They didn't seem to though.