SAP - You Should've Caught Hasso's Keynote

                 SAP, SaaS and Memory Resident Databases

Some time ago, several of the Enterprise Irregulars got an A1S briefing from Dr. Hasso Plattner, board member and founder of SAP. A1S was the project name for an all-new solution SAP was developing for the SMB marketplace. That product was introduced last Fall as SAP Business ByDesign.   

 

In my conversations with Dr. Plattner back then, we discussed whether they would utilize a memory-resident database as part of the Business ByDesign product. I was told then that they would and we had a spirited discussion about the value of such an approach. Back then, I should have been clearer and asked when this technology would be brought to bear. Now we know – it’s now. 

 

 

Memory-resident databases are immensely valuable as they reduce search times by several form factors. There is no delay/latency waiting for a disk drive to re-position its read/write head or waiting for the drive to rotate to the correct spot. Memory resident databases don’t suffer the performance hits that conventional media do when data is fragmented across numerous parts of a drive. 

 

To see how effective a memory-resident database works, SAP executives showed how quickly a 1.3 billion record database of Point-of-Sale data is searched. In a live demonstration this morning at SAPPHIRE, sub-second results were consistently achieved. 

 

 

Workday has been using in-core or memory-resident database architecture for some time now. 

 

SAP has been both experimenting with it and now, with the Business Objects acquisition, is ready to really exploit it for both Business ByDesign product users and beyond. Several times today, Dr. Plattner remarked how the product could be used by Wal-Mart to parse billions of customer transactions.

 

The new architecture works as follows (see photo):

• A user initiates a transaction
• The system wants to add/delete/update a record
• The software updates the memory-resident database, the MaxDB rdbms and BI data stores
• When the user wants to review, search, report, etc. this data, the actions are performed in-memory at a form factor faster time

Latency in high-technology is rarely desirable or valued. In a SaaS (software as a service) environment, latency is a real problem as users could be anywhere, connecting via the Internet on spotty connections accessing faraway servers. SAP developers are being ordered to produce rendered screen pages in 300 milliseconds or less. Given all of the ways that page loading can be slowed down (e.g., graphics-intensive images, inefficient searches, etc.), anything that helps make SaaS fast is a godsend. 

 

SAP intends to bring this technology to its other product lines beyond Business ByDesign. The initial catalyst will be Business Objects and the T-Rex architecture. Last year, prior to the Business Objects acquisition, SAP was able to search through 36 million accounting transactions/second using memory-resident database technology. Now, the company can easily search over 1 billion records/second.

As we’ve written previously re: Workday’s approach, the use of an in-core database obviates the need for subledgers and other redundant data stores. Dr. Plattner indicated that the degree of ‘compression’ is 10-20X.   

 

Dr. Plattner also indicated that graduate students at the University of Potsdam  are working aggressively in this area with SAP.

 

The implications of this technology decision are:

-          RDBMS technology will get relegated to a backup media or light-duty technology

-          Users will be able to perform previously unimaginable searches, reports, analyses, etc.

-          Older application software products will need re-working. To stay RDBMS-only focused will become competitively untenable.

-          Universities, technical schools, etc. need to develop more skills in the areas that memory-resident technologies open up (e.g., analytics, competitive intelligence)

-          That we all need to buy more stock in solid-state memory makers like Spansion.

Comments

[Nitin Goyal]

Oracle has a product - Timesten - http://www.oracle.com/timesten/index.html

This was acquired by Oracle in 2005 and does the same stuff - in memory processing of data.

I would not hazard a guess to say that RDBMS technology will become relegated to backup/light duty processing. There are still many challenges in using Memory based products and unless they are backed by a robust database engine, they are pretty much worthless.