Alexander Zeier, co-designer of SAP's HANA appliance, speaks to Computer Weekly about the evolution of in-memory analytics.
Zeier joined Accenture eight weeks ago, having spent the last few years at MIT (Massachusetts Institute of Technology) where he was a visiting professor in real time analytics, predictive analytics, forecasting and big data.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
But his claim to fame is the time he spent at SAP where he worked alongside SAP co-founder Hasso Plattner, to design the first working version of HANA, SAP's in-memory database appliance.
Zeier also holds 10 patents related to in-memory technology and applications for enterprise systems and is co-author of In-Memory Data Management.
Zeier worked with some of SAP's largest customers to build the HANA database appliance.
He says: “I built a prototype with Hasso for Colgate, which had a standard SAP system. When connected to the company's relational database, it processed 280m invoices in 20 minutes. With HANA the same transaction took 1.5 seconds.”
New analytical applications made possible with in-memory databases
- Smart meter reading
- Automatic purchasing based on statistical analysis
- Determine a product's popularity by analysing the supply chain management system
- Analyse sales data in real time
- Improve application usability by querying the live transactional system
- Instant response time on mobile applications that cross reference ERP systems
Source: In-Memory Data Management, by Hasso Plattner and Alexander Zeier
The prototype was built over five years ago. Today, he says: “In 95-99% of the time, HANA responds in under a second.”
Part of the improvement has come about due to the work Intel has done to improve data throughput between memory and the processor cores. When HANA was first built, the speed of moving data between memory and the processor core peaked at 5-6 Gbps.
“Intel has now increased data throughput to 100Gbps,” says Zeier. The result is a 20x increase in data throughput, which means HANA can process more data, leading to faster query times.
He says HANA also uses parallel programming to enable columns of data to be processed at the same time. So as more processor cores are added, the speed of HANA's query processing accelerates.
Zeier says each processor core can process 2MB of data per millisecond. And given that each blade within the hardware configuration can be use 80 processor cores, Zeier believes HANA has a huge amount of headroom to scale.
So what can a business do with all this power? Zeier believes it can be harnessed to power new and innovative business processes.
“The speed of thought is 700ms so you can build really cool mobile applications with instant response times.”
HANA can essentially keep up with the human brain, which means that applications are perceived as being instantaneous. “You can create new business processes based on HANA,” says Zeier.
For instance, it is now possible for a financial institute to run risk profiling in real time.
The arguments against in-memory databases
Some people would argue that in-memory databases like HANA are only suitable for the largest companies with the biggest data and the fattest wallets. Zeier disagrees.
“You can now buy 1TB of memory for $100,000,” he says, making the HANA appliance relatively low-cost in terms of datacentre computing.
Another criticism of in-memory database appliances is that they are proprietary.
This is not the case with HANA, according to Zeier. The hardware architecture is based on standard Intel x86 processors. Cisco, IBM, Hitachi Data Systems and HP are among the hardware partners providing HANA systems.
The underlying operating system (OS) is SuSE Enterprise 11 Linux. But to improve performance, Zeier says HANA includes low-level programming that is optimised for the level 1, 2 and 3 caches on the Intel processor.
The next age of IT
Business computing began on the mainframe, then shifted to client server computing.
Zeier believes in-memory processing represents the third era of computing. If SAP R/2 was of the mainframe era and R/3 represented SAP's client server architecture, for Zeier HANA could be regarded as R/4.
And from what SAP has been doing in Madrid at its TechEd conference, it seems the supplier is certainly pushing its in-memory database appliance to power core enterprise applications. According to Zeier, “In 2006, when we developed HANA, it was as a single data source for transactional and analytical systems.”
While SAP has previously positioned HANA mainly for accelerating business analytics, with SAP 360 CRM, the company has returned to Plattner and Zeier's original vision. HANA is now a transactional and analytical database appliance.