SAP announced last week the availability of SAP BusinessSuite powered by Hana, fulfilling a vision it laid out three years ago to leverage the power of in-memory processing not only for analytics but also for transaction processing.
I'll confess that I was one of those skeptics about the OLTP aspect of SAP's in-memory vision. At the time, in-memory was gaining ground in the BI space, with momentum mainly by smaller vendors such as QlikTech, TIBCO Spotfire, TM1 (acquired by Cognos then IBM). In the mainstream BI camp, SAP had its BW Accelerator, MicroStrategy added in-memory to its OLAP Option, and Microsoft had just released the PowerPivot, in-memory solution for Microsoft Analysis Services, but in 2010, none of these solutions were mainstream. As well, in-memory solutions for OLTP such as TimesTen (acquired by Oracle in 2005) and SolidDB (acquired by IBM in 2007) had little marketshare or momentum.
From the ERP viewpoint, SAP Chairman and cofounder Hasso Plattner claims the need for speed in the ERP system is being driven largely by mobile, where end customer apps demand immediate responses. Here, the use of an in-memory system instead of a relational database has the potential to change business processes as well as lower the cost of ownership. SAP demonstrated the use of BusinessSuite on Hana for material resource planning (MRP) in which a planning analyst must consider existing inventory, supply, and demand. Using a conventional database such processes run in batch, typically overnight depending on the data volumes and required data inputs. With Hana, an MRP forecast can be completed in seconds. If you take the business example of say the current flu epidemic, the value of doing this real-time could potentially solve the problem of insufficient flu medications (in thin supply at our local pharmacy, and yes I got my flu shot but still got the flu!) more immediately.
Customer John Deere has been using Hana for analytics for just over a year. With the success of some proof of concepts, they tested Hana for ERP use as a way of helping their customers and bringing new products to market. Derek Dyer, Direct or Global SAP Services at Deere, says Hana will simplify the IT stack and lower the cost of ownership.
Hasso also argued that the use of Hana as an OLTP database way brings OLTP reporting back to the OLTP. How and where to do operational reporting has been a big point of confusion and frustration for many BI teams and operational report developers, both because of issues in the OLTP and capabilities of the BI tools. Nobody wants to risk running an operational report that cripples data inputs, but when real-time data is needed, you don't want that report to hit potentially stale data in the data warehouse. In this regard, I was glad that Hasso took a pragmatic approach (in contrast to some big data vendors) of conceding that the data warehouse is still necessary for data cleansing and transformation and bringing disparate data sources together.
From the strictly BI perspective, SAP could be more clear about Hana as a platform for BI. Few in the industry realize, for example, that Web Intelligence can access Hana as a data source, as of SAP BusinessObjects 4.0 Feature Pack 3 (released in June 2012). This did at least get mention at the press conference, but it has rarely been touted in the past as Explorer and Visual Intelligence on Hana have gotten most of the attention. While Visual Intelligence and visual data discovery in general have been the buzz of late, Web Intelligence is the bread and butter of the BI platform, both widely deployed and key to self-service BI. Most often, Web Intelligence queries hit relational data warehouses or BW. Hana as an alternative to a slow data warehouse or data mart seems not be to a key part of SAP’s strategy, unless of course you are already a BW customer.
Also not addressed is how Hana integrates with existing relational databases, because at some point customers may want more granular data or historical data on disk. This seems to be a weakness in Hana compared to Oracle Exalytics, IBM Cognos 10.2, and MicroStrategy all of which who have software advisors to recommend what to leave on disk and what to put in memory. More recently, both QlikView and TIBCO Spotfire latest releases have included drill through from in-memory to in-database options. So, for example, in TIBCO Spotfire, if a user is exploring summary data that has been loaded into Spotfire in-memory and then needs additional details that are stored in the Teradata database, the user can interactively and seamlessly fetch more data as they explore and drill. With Hana, the data seems to be preloaded to memory or not, so accessible or not. Perhaps this is by intention as co-CEO Jim Snabe previously dismissed the need for a hybrid solution, saying they don’t last. I will agree that Hybrid OLAP never gained acceptance, but at the very least, the in-memory advisor approach in competitive tools seems to provide a transition strategy.
For a side-by-side comparison of leading vendor's OLAP and in-memory approaches, see the latest BI Scorecard Product summary.
Cindi Howson, BI Scorecard