SAP CIO predicts in-memory HANA will replace relational databases

But will companies still retain an archived disk?

Software giant SAP is likely to move away from relational databases in the future and focus on its in-memory offering, HANA, according to SAP's CIO, Oliver Bussmann, who spoke to Computing at the SAP User Group Conference this week in Birmingham.

HANA uses in-memory software to analyse huge volumes of data in real time.

"My prediction is that in the next five to ten years relational databases will disappear and all SAP products will be using HANA," said Bussmann.

"In-memory allows companies to improve the speed of their data analytics by a factor of 100,000 [compared with relational databases]. This means that analytics that took hours can now take a second."

SAP began looking at in-memory capabilities three years ago and started developing the technology in early 2010.

Bussmann's team conducted a proof of concept in October of 2010, which saw SAP's data analytics improve by a factor of 14,000: a request now takes one second where it previously took five hours.

SAP then delivered a roadmap for companies to utilise HANA, and 100 of its customers have now implemented the technology.

"The first step is for companies to put the HANA appliance alongside an ERP system. Then you should extract as much data as possible, put it inside the HANA appliance, and run analytics on top of that in real time," said Bussmann.

"Companies should build applications that are 100 per cent optimised on the HANA software. There are many examples of this already, and these include liquidity and risk-planning for banks."

Bussmann argues that many more companies are now looking at in-memory data analytics because the technology has become cheaper in recent years.

"We realised that memory has become very inexpensive. You can get a 2 terabyte machine for £85,000, with 64 processers, so it is now pretty cheap to own the hardware," he said.

"However, you need software to optimise the hardware, which HANA does. If you move data into the main memory, you will have a factor-of-ten performance improvement. But if you leverage multiple core processors and get the software to spread the work efficiently, you see more drastic improvements in speed."

Despite the optimism and the fact that SAP now has more than 100 customers using HANA, Bussmann still believes that there is room for improvement with the product.

"We need to improve the modelling of the data - we need to make this more stable," said Bussmann.

"We also need to scale up our development for new applications and allow the ecosystem to establish itself. These things take time."

SAP CIO predicts in-memory HANA will replace relational databases

But will companies still retain an archived disk?

Mike Thompson, principal analyst at Ovum, argued that SAP's prediction is slightly inaccurate: he claimed in-memory databases will not replace relational databases, but rather on-disk databases.

Thompson suggested it is not the relational structure of the data that causes latency problems; it is the fact that queries to conduct analysis are being done on disk.

"I would suggest that what SAP really means is that it is going to replace on-disk databases. A relational database allows people to write a query to it in SQL language and problems arise because the data is on disk," said Thompson.

"If you send the database a query, a lot of mechanical interaction has to take place. It's not the structure of the data that is the issue here; it is the physical movement and the mechanical element of an on-disk database.

"It takes time to make a query in this way. The relational model is as good as we have at the moment, and most companies have a relational database."

The term "relational" simply refers to how the data is structured on the disk: it allows organisations to sort and access data quickly.

"An in-memory database takes all that structure and data, and puts it into a CPU, into memory," said Thompson.

"If it gets a query, it doesn't have to submit it to the physical disk somewhere, and it doesn't have to work out where that data is. It knows where it is because it is sits right next to it."

Thompson also argued that companies are still likely to use disk to store their data in the long term, although in-memory may become more popular.

"One of the issues with in-memory databases in the past was that they were very good to bring the data out and work on it, but all the data would be lost if there was an outage. So it would have to be relayed to disk at some point," said Thompson.

"However, chips now allow companies to create snapshots, and you can roll back to the point before where the data was lost. So theoretically, you don't now need a disk.

"Practically, however, most companies will retain their disks and utilise multiple in-memory databases. They will alternate in-memory databases. When one is not being used, it will relay data back to an archived disk."