Government // Enterprise Architecture
Commentary
4/11/2012
05:09 PM
Josh Greenbaum
Josh Greenbaum
Commentary
Connect Directly
LinkedIn
Google+
Twitter
RSS
E-Mail
50%
50%

SAP Says Hana Tests Well On Performance, Scalability

If SAP's new benchmarks are accurate, in-memory technology could take scalability concerns off the table and help SAP compete with Oracle on better terms.

SAP has been trying hard to demonstrate the business case for its in-memory, RDBMS-killer--Hana. A key issue has been proving both Hana’s scalability and appropriateness for serving as an analytical platform for large SAP transactional systems. The case for both of these proof points appears to have been significantly advanced by the results of recent internal benchmarks on Hana that SAP shared with me this week.

The results are unverified and therefore must be viewed in light of the on-going Oracle-SAP marketing war. But at a minimum, the data puts the onus on Oracle to match or better, if it can, what look to be relatively impressive numbers.

Here’s a summary of this latest Hana scalability test: 100 billion records of relational data, representing five year's worth of SAP sales and distribution data, were loaded into Hana, which compressed the original 100 terabytes of disk data down to 3.8 terabytes of RAM. The Hana system was running on a commercially available IBM X5 16-node cluster, which is four times as large as the platform used for SAP’s previous Hana scalability test. The cost of such a system would be around $600,000, SAP says.

[ Want to read more from Josh Greenbaum on SAP? See SAP's Success And Looming Challenges. ]

What's impressive is that SAP is claiming that not only does this configuration deliver sub-second response times--300 to 500 milliseconds for a set of standard SAP Business Warehouse-like queries--but that this is the identical query response time it got when it ran Hana on the smaller 4-node X5 cluster. Equally impressive is that these response times were achieved with no database tuning; the data were partitioned at load time, but nothing in the way of typical RDBMS tuning tricks were used to achieve these results. The test also included ad hoc (as opposed to BW-like) analyses that compared customer data across different time periods with equally impressive results. Being able to do this without using material views is another potential feather in Hana’s cap.

The benefits don’t just accrue to analysts looking to leverage data in the SAP Business Warehouse or a generic data warehouse. SAP is also counting on a tool called Landscape Transformation, which replicates data from the SAP transactional system to Hana, to let customers analyze real-time transactional data alongside their BW data using Hana.

There could be two possible and important effects from the results of these scalability tests, once they're independently verified. The first would take the Hana scalability issue off the table and drop it squarely in the competition’s court. And the second is that these results make the business case for Hana center on relative functionality and price/performance, making it a potential winner for both the business user and the IT department.

I’ve placed the caveats around this data for an important reason: SAP needs to have these results independently reviewed, and, once reviewed, Oracle and any other competitor that feels threatened by Hana should be given the opportunity to respond with scalability data of their own. If SAP’s data holds up under independent scrutiny, it will shift the competitive conversation in SAP’s favor. And that, in turn, will be further proof of the waning importance of relational technology as a cost-effective and functionally superior approach to modern analytics.

Josh Greenbaum is principal of Enterprise Applications Consulting, a Berkeley, Calif., firm that consults with end-user companies and enterprise software vendors large and small. Clients have included Microsoft, Oracle, SAP, and other firms that are sometimes analyzed in his columns. Write him at josh@eaconsult.com.

InformationWeek is conducting a survey on the state of enterprise applications and business processes and organizations' priorities in evolving technology over the next 12 to 24 months. Upon completion of our survey, you will be eligible to enter a drawing to receive an 32-GB Apple iPod Touch. Take our Enterprise Applications Survey now. Survey ends April 20.

Comment  | 
Print  | 
More Insights
Comments
Newest First  |  Oldest First  |  Threaded View
Sam Iam
50%
50%
Sam Iam,
User Rank: Apprentice
4/16/2012 | 6:36:51 AM
re: SAP Says Hana Tests Well On Performance, Scalability
I think the real concern for people considering HANA for future RDBMS is reliability and resiliency. If you lose a server or have a memory error (which is not uncommon), you lose all of that data because there is no persistent state in DRAM. This is not a major problem for OLAP, data warehouse workloads because you can just reload the read only batch, but it is a real problem for OLTP workloads which constantly have incoming I/O. It is not as though the database would be gone permanently, but you would have to reload it from some snap copy on hard disk, ensure that the data is synchronized, and then restart it which would mean downtime. Maybe this can be resolved by clustering and mirroring data around a bunch of servers in a Hadoop style configuration, but, if that is possible, I haven't heard anything about it.

A second issue for HANA is that SAP writes about near real time responses as though they always hugely beneficial. In some trading application or massive analytics application, it will be an advantage. For the majority of OLTP workloads, it is not clear that anyone would achieve some unique business advantage by being able to get to a SAP general ledger record or benefits enrollment screen half a second faster than before. If the price is the same or less, then you might as well have faster response times. If that $600,000 number is correct, that would probably be lower than the cost to buy a tier one 100 TB SAN alone. IBM's X5 gear is ideal for this workload with the IBM memory expansion technology, so it might be possible.
Register for InformationWeek Newsletters
White Papers
Current Issue
InformationWeek Tech Digest September 18, 2014
Enterprise social network success starts and ends with integration. Here's how to finally make collaboration click.
Flash Poll
Video
Slideshows
Twitter Feed
InformationWeek Radio
Archived InformationWeek Radio
The weekly wrap-up of the top stories from InformationWeek.com this week.
Sponsored Live Streaming Video
Everything You've Been Told About Mobility Is Wrong
Attend this video symposium with Sean Wisdom, Global Director of Mobility Solutions, and learn about how you can harness powerful new products to mobilize your business potential.