In the review of my book (Oil ITJ June 2011), you question the linkage between the in memory database (IMDB) and various seemingly unrelated topics—let me explain.
The Microsoft Surface device was mentioned because of its capability to share information across a group of users. For this to happen effectively, queries have to be answered within the human attention time span—i.e. with a maximum acceptable delay of about 8 seconds. Such a response time can only be achieved with a database of this type.
We also referred to the ‘cloud’ since the best way to run an on-demand application in the cloud is to use an in-memory database. Here, columnar data storage gives us 5-20x compression factor for customer data. Multi-tenant use makes for very large tables and here again, speed is everything. The columnar database’s schema can be changed on the fly, adding new fields and simplifying upgrades.
On the hardware side, we highlighted the Intel Nehalem architecture as the current leader in terms of CPUs per board, cores per CPU and RAM. A database server with 4,000 cores on 100 nodes with 50 terabytes of RAM is currently operating in San Jose at a sensationally low price point!
While everybody talks about virtualization, with the IMDB, there is no need for it. Performance is significantly higher running natively on the operating system.
Regarding big data, our experience with SAP Hana running real customer scenarios is sensational. We have replaced custom applications built with Oracle tools running on an Oracle database with the same application on Hana and seen huge run time improvements. 1,000x is normal, 10,000 frequent, and on 3 occasions we achieved a 250,000x speed up. Now many applications can run on big data as a transaction—returning sub 8 second query results.
The book is intended to help students understand a new technology which will change enterprise computing profoundly. Any marketing flavor was unintended and is a mistake that the authors let slip in. There is definitely no FUD whatsoever. In fact the reality is even better than what we anticipated two years ago.
Best regards,
Hasso Plattner,
Hasso Plattner Institute,
Potsdam.
This article originally appeared in Oil IT Journal 2012 Issue # 9.
For more information or to comment on this topic email here.