Hi, I'm studying the OLAP technology. I thought that I understood how
this technology work, until i found the POLA Server. I understand that
this server build the cube and aggregation in the Memory. I think that
in this way isn't possible manage a big quantity of dates.
All MOLAP server work in this way?

Some OLAP servers keep all the structure and data in memory. You can
look for one product that is more memory/cpu/storage efficient which is
HyperRoll (http://www.hyperroll.com)


Thanks,
LK

FM2 wrote:

Those that perform everything in memory are the easiest to write, and
generally the fastest. However, the disadvantage is that there are size
limits (as you've pointed out), and the results are volatile (i.e. data is
lost if server program or server machine crashes).

There are several MOLAP servers that store data on disk, and some that even
implement a multi-level storage architecture. For instance, the old Holos
product could store aggregations, or ad hoc data changes for "what if"
analysis, in a separate sparse memory storage level. This conceptually sat
above a main disk (or SQL) lower storage level. The overall cube, as
utilised and reported on, would appear as the combination of these two
layers -- anything in the top level hiding anything in the bottom level.
This also facilitated full transaction support (i.e. commit/rollback).

Tony Proctor

"FM2" <NOSPAM (AT) NOSPAM (DOT) it> wrote

....hence the adjective "old" in my post Jörg. However, I would say that it's
still interesting from an historical point of view, for anyone studying
implementations, since it was fundamentally different to other OLAP
products, both then and now.

Tony Proctor

"Jörg Narr" <joerg_narr (AT) swissonline (DOT) ch> wrote

The amount of data allowed using Applix TM1 with a 64 Bit server
technology is almost limitless. TM1 is also a memory based OLAP
technology. The statement that memory based applications can't handle
the quantity of data is now misnomer.

FM2 wrote:

But there is a reason why most commercial databases are not primarily
memory resident. There's truly nothing from stopping you from
configuring virtual disks if you've got a ton of memory but most people
live on a budget. Any product that uses only memory is limited - and if
you can get ACID for them, fine. But...


Rick wrote:

That is the reason why products like HyperRoll utilize the fine balance
between memory and disk to speed up query time compared to other old
legacy systems.