SAP may have stumbled onto an Oracle killer: in-memory technology that could, in theory, quash the need for a relational database in some cases.
eWEEK has learned that SAP has sussed out a way to organize its business intelligence data in columns versus tables, storing and indexing the data in memory and then running it all on blade servers.
The result is faster queries than would be possible by tapping data stored in a data warehouse or relational database. And with the cost of memory plummeting over the past few years, SAP executives say the Walldorf, Germany, companys in-memory technology is a much cheaper data storage alternative to traditional databases—for its BI customers, that is.
But eWEEK has also learned that SAP is working on in-memory data management capabilities that could go beyond BI to other areas of the application stack, replacing the need for a relational database in new software installations.
With about 55 percent of SAP implementations sitting on Oracle databases, theres a potential for disruption. That said, the potential of in-memory technology isnt lost on Oracle; it bought in-memory database provider TimesTen last year.
"What we are seeing with text search and Google … [is] showing us the way of using main memory for organizing text data," said Vishal Sikka, SAPs chief software architect. "Weve all used main memory in the past. … Now, in the case of analytics or unstructured search, its become flexible enough to do … application-specific data management."
During its Sapphire user conference in Boston in May, Shai Agassi, SAPs head of software development, demonstrated the companys in-memory technology, now renamed the BI accelerator, an analytical engine within SAP NetWeaver Business Intelligence.
SAP put its technology on IBM and Hewlett-Packard servers to speed up, by orders of magnitude, querying capabilities. It put the boxes out in the field at some big companies with pretty large data warehousing needs: Coca-Cola, Whirlpool, British Petroleum and Novartis. The results, according to Agassi, were astonishing: a 90 percent increase in reporting performance, with queries cut from 60 seconds down to 3 seconds in the case of Coca-Cola.
"We knew in the lab this was beyond cool, but we didnt know how much impact there would be on the day-to-day life of users of data warehouses," said Agassi.
SAPs use of in-memory technology brings up an interesting question: If the technology could potentially disrupt Oracles database business, wouldnt it also disrupt IBM and Microsoft as well? Given that SAP is partnering with both companies, it will be a delicate balance to strike.
Which leads to yet another question: Can SAP dovetail its use of in-memory application technology with the burgeoning use of in-memory database technology—and put a hurting on Oracle at the same time?
The answers may be a couple of years out, but theres no doubt SAP and one of its key database partners, IBM, are considering the options.
"IBM is definitely spending resources and time tuning its DB2 database specifically to run with SAP," said Donald Feinberg, an analyst at Gartner, in Stamford, Conn. "As for in-memory, anything they can do to speed things up, [IBM and SAP] will do … together."
In-memory database technology keeps data in memory, rather than stored on disk. The technology is now used primarily in financial and telecommunications applications, where speed is critical, but people in the industry suggest it could be relevant in SOA (service-oriented architecture), RFID (radio-frequency identification), manufacturing and e-commerce as well.
Much transactional data from enterprise applications is now stored in relational databases. That data could be moved to in-memory for huge gains, said Joshua Greenbaum, principal at Enterprise Applications Consulting, in Berkeley, Calif.
"The in-memory database could … disrupt the top tier of database vendors: Oracle, Microsoft, IBM. Thats because [when] you start moving to in-memory, you potentially remove the requirements for this huge infrastructure."
Greenbaum and others have pointed out that in-memory database technology is not necessarily a replacement for standard relational databases—but for new applications and installations, it can be.
That said, Oracle is a formidable competitor. At this point, its the only major database company with in-memory capabilities, gained with its 2005 buy of TimesTen. Its conceivable that Oracle could also be looking at using TimesTen technology to optimize querying across its applications. It has already front-ended the TimesTen in-memory database with its Oracle Database 10g.
IBM, however, has recently formed a relationship with Ants Software, an in-memory database provider based in Burlingame, Calif. (Ants also has a relationship with SAP, but both Sikka and Ants CEO Boyd Pearce declined to comment on it.)
In a May 5 note to investors, Ants officials detailed the companys relationship with IBM on its Web site: The two companies are making sales calls together to a number of large enterprises (including the U.S. Army) and are putting forth a business case for cost savings through database consolidation into the Ants Data Server with IBM providing technical support.
"The reason IBM is talking to us is, if you dont have an in-memory technology—and Oracle does for some applications—that may influence a whole lot of spending decisions," said Pearce.
"It could mean the decision—a huge one—on which database you use. If IBM doesnt step up to having a comparable solution to TimesTen, theyre going to lose out."
IBM and Ants are working on several initiatives, according to Pearce. "One of our configurations with IBM is we add an Ants [server] to DB2 so you can cache important information and have the in-memory component and can make the whole thing go faster," he said.
The end goal is to make DB2 much more accessible and popular so that solutions can be consolidated around DB2 and users can get rid of their other databases, according to Pearce. "Our goal is to be able to run any applications on a combination of Ants and DB2 without having to change the applications at all," he said.
IBM currently has some in-memory-like capabilities in DB2 9, the latest version of its database software that has, coincidentally, been optimized for SAPs applications.
"There are a few different uses of memory with our standard database," said Peter Kohlmann, DB2 product manager at IBMs DB2 labs, in Toronto.
"For example, we have buffer pools that keep as much information in memory as is reasonable—it winds up on disk at some point. What weve moved to in 9 is self-tuning memory management that manages the size of key memory parameters, which means we will optimize what stays in the buffer pool."
Kohlmann said IBM does not have a "hard requirement" from SAP in terms of in-memory database technology. But Bill McDermott, CEO of SAPs North American arm, confirmed that the two companies are working together to further the work done with the BI accelerator.
"[In-memory] is definitely something were working on, and in partnership with IBM," said McDermott in Newtown Square, Pa.
SAP is reportedly working to upend Oracles application intentions by attacking the companys database stronghold. The weapon: in-memory technology that could ride shotgun with applications. The lowdown on in-memory databases:
WHAT THEYVE GOT
- Infrastructure software designed for low latency, high- volume data, and events and transaction management
- Optimized for deployment within the application tier
- Maintain data in memory rather than on disk
- Fast query capabilities, since data is not locked on disk
- Used now in telecommunications and financial markets for high-volume, real-time data tasks
- Increasingly used for midtier data management in SOA infrastructures
WHAT THEY DONT HAVE
- Storage for raw data, other than pushing data back to a data warehouse or relational database
WHOS DOING IT
- Oracle TimesTen
- Ants Software
- McObject (ExtremeDB)
- Other niche vendors
Source: eWEEK reporting, Oracle TimesTen Web page