Government // Enterprise Architecture
Commentary
5/1/2012
01:24 PM
Doug Henschen
Doug Henschen
Commentary
Connect Directly
Google+
LinkedIn
Twitter
RSS
E-Mail
50%
50%

Hana and Exalytics: SAP's Hype Versus Oracle's FUD

Getting to the facts in the war of words surrounding SAP's Hana and Oracle's Exalytics platforms.

At one point Kurian rattled off a litany of claims about Hana limitations that would make a would-be customer's head spin. Let's lay a few of those to rest.

* Hana does support SQL and MDX.

* Hana does support parallel query execution. In fact, it supports massively parallel query execution.

* Hana does not support a bunch of stuff related to ROLAP and MOLAP--like indexes, aggregates, and materialized views--because it does away with those artifacts entirely. It makes calculations instantly on the fly--using the latest data and all available detail. SAP's analogy: Hana doesn't need a MOLAP hay loft or a materialized view whip because it's not a horse and buggy; it's an automobile.

* Hana does support unstructured data analysis. In fact, the database's origins were in columnar text processing, and SAP BusinessObjects has since added text-analysis capabilities that can be used in conjunction with Hana.

[ Want more on in-memory analysis needs? Read SAP And Oracle: Get Real About In-Memory Analysis. ]

One key point by Kurian that IT leaders need to consider: "If you have a data warehouse or mart in which only a small amount of the data is actually hot, why would you put it all in an in-memory-based system?"

I've asked SAP that same question more than once and have been told that:

A. DRAM is getting so cheap it will inevitably replace disk, and

B. that if the archives are truly large, Hana will work with warehouse-oriented products such as Sybase IQ.

However, if you still need a separate warehouse, it kind of defeats the argument that you can do away with extra database licenses and layers of unnecessary infrastructure. Until disk-based memory is truly obsolete, customers will have to consider what's hot and what's not and ask tough questions about what's required for the in-memory performance they're after.

On the subject of Hana's cost, Oracle's apples-to-apples comparisons to Exalytics aren't appropriate because the data footprints and database license and hardware requirements would be vastly different even for the same customer. With one copy of data with Hana instead of two or three with Oracle--plus the elimination of redundant aggregates, indexes and materialized views--Hana users could expect to have a much smaller data footprint. The average reduction won't be known until the market sees more BW-on-Hana deployments, which will take some time. As for applications on Hana, well, we'll have to wait until that becomes a reality.

Getting To The Bottom Line

Pricing Exalytics in a small data-mart scenario at around 500 gigabytes, Kurian said the hardware would be $135,000 and the TimesTen database another $690,000 for a total of $825,000. (He did not mention Oracle BI Foundation software, which is also required.)

By comparison, Kurian claimed SAP's cost would be $362,000 for hardware and $3.7 million for software. SAP's Steve Lucas says Hana's cost in this scenario, including hardware and software, would be $500,000, but by my calculation, using BW-on-Hana list prices (of $79,000 per 64-gigabyte unit, as reported here ) and a 50% database overhead allowance (which SAP calls for), the software cost alone would be north of $1.2 million. SAP must be counting discounts and incentives it's throwing in to spur sales.

The bottom line is that Oracle's claim that Hana costs 5-to-50-times more than Exalytics is exaggerated--in large part because it's based on same-size deployments, when Hana will allow smaller deployments. But it's also hard to believe SAP's sweeping statements about the affordability of DRAM-based systems and Hana overall.

"The reason why new applications have been emphasized is that Hana is expensive," says Gartner's Feinberg. "So you need an application that's going to give you strong business value."

That customers can now run BW and Planning and Consolidation on Hana brings more value to the equation. But customers won't see the full potential of Hana until it's running the warehouse, the core transactional applications, and a breakthrough, business-value-driving new application.

Previous
3 of 3
Next
Comment  | 
Print  | 
More Insights
Comments
Newest First  |  Oldest First  |  Threaded View
UliBethke
50%
50%
UliBethke,
User Rank: Apprentice
12/24/2012 | 9:17:55 AM
re: Hana and Exalytics: SAP's Hype Versus Oracle's FUD
SAP Hana can't really be compared to either Exalytics or Exadata. Oracle has no similar product that can currently compete.

For my full two cents have a look on my blog: http://www.business-intelligen...

brian9p
50%
50%
brian9p,
User Rank: Apprentice
5/31/2012 | 4:17:42 AM
re: Hana and Exalytics: SAP's Hype Versus Oracle's FUD
SAP has a plan to migrate its large base of application users to its database platform. It already has the fastest growing share in the database market and is eyeing the number two position by 2013 http://www.informationweek.in/...
Amyn Rajan
50%
50%
Amyn Rajan,
User Rank: Apprentice
5/8/2012 | 5:36:29 PM
re: Hana and Exalytics: SAP's Hype Versus Oracle's FUD
HANA does support the MDX query language. In fact, Simba worked very closely with SAP to build MDX capabilities directly into HANA. Uwe Fischer from SAP and I presented at the SAP BI conference in March this year how you can easily connect from Microsoft Excel to SAP HANA using OLE DB for OLAP (ODBO) and the MDX query language.

From the Oracle side, the Oracle database does not support the MDX query language natively like HANA does. For Oracle database OLAP Option, you need to separately license the Simba MDX Provider for Oracle OLAP to get this capability.

Oracle Times Ten also does not support the MDX query language.

Oracle Essbase does support the MDX query language. However, Essbase only supports XMLA and does not support ODBO. Therefore, many applications like Microsoft Excel will not be able to connect to Essbase using the MDX query language.

To summarize:

SAP BW - MDX (yes), ODBO (yes), XMLA (yes)
SAP HANA - MDX (yes), ODBO (yes), XMLA (yes)
Oracle database OLAP Option - MDX (add-on from Simba), ODBO (add-on from Simba), XMLA (coming soon from Simba)
Oracle Times Ten - MDX (no), ODBO (no), XMLA (no)
Oracle Essbase - MDX (yes), ODBO (no), XMLA (yes)
Sam Iam
50%
50%
Sam Iam,
User Rank: Apprentice
5/2/2012 | 8:11:44 PM
re: Hana and Exalytics: SAP's Hype Versus Oracle's FUD
Exactly, they will need some sort of redundant clustering configuration before anyone uses HANA as an RDB. If it is just a stand alone x86 server with a ton of DRAM, the chances of failure are way too high for a production DB. Even if the process of reloading data from HDD is simple (and there are no compression, dedup, transformation processes), it is going to take awhile for HANA to reload several TB of data from HDD to any location. You are back to the HDD IOPS bottleneck issue.

Another issue that concerns people about HANA is that SAP seems to be pursuing an Oracle "own the stack" strategy. Originally they were an ERP provider, then they added Netweaver and became and application server and pseudo middleware provider, then they added BI and DW, now they have Sybase and HANA and are becoming a database provider. They say that they will be an open provider, preserve choice, etc, but they will have a great deal of power over accounts that are SAP from DB through applications.
Sam Iam
50%
50%
Sam Iam,
User Rank: Apprentice
5/2/2012 | 7:50:27 PM
re: Hana and Exalytics: SAP's Hype Versus Oracle's FUD
Where am I supposed to "do my home work"? SAP has zero implementations of HANA as RDB, it's not even supported at this point. There is no documentation on HANA for RDB.

"This is real stuff. HANA simply reads the data from disks on startup that is intitially required in RAM."

I get it. How long does it take to rebuild a few TB from disk to DRAM in the case of a memory error or server failure? What happens to all of those IOs that occur in a live OLTP database when HANA is not online? How does it sync the HDD copy to the production DRAM instance to enable near real time fail over so your primary RDB is not down for an extended period? They will have to answer those questions before anyone puts a live SAP RDB on HANA.

"Businesses are screaming for realtime access to OTLP data. It has always been so problematic and the pain runs deep."

I am familiar with the IOPS bottleneck coming off of HDD. My question is, does closing this bottleneck and improving OLTP response time "transform" anyone's business, especially SAP's core industrial and retail accounts? If you can get an materials config processed at GE half second faster or process a POS transaction at Wal-Mart half a second faster, what are the concrete cost savings or new revenue opportunities? The situations where it might be transformational are in the industry where SAP does not have a strong account base, financial services.

"Data latency is problematic and data duplication baloons the storage cost of traditional warehouses"

Are you talking about databases or data warehouses? Data latencies are generally not an issue in data warehouses. Various in-memory, SSD, massive parallel processing and columnar compression technologies have been around for years. You can make a data warehouse perform at insane rates with something like a Netezza appliance or TM1 for small workloads capable of being run entirely in DRAM.

It is not as though SAP does not use dedup/compression and you do not pay for it. It is baked into HANA instead of being a separate appliance.

This could be as good as SAP thinks it is going to be, but at this point they have an in-memory DW. I will await the details of the RDB.

D. Henschen
50%
50%
D. Henschen,
User Rank: Author
5/2/2012 | 4:06:41 PM
re: Hana and Exalytics: SAP's Hype Versus Oracle's FUD
Good comments. Indeed the current Hana platform includes backup to SSD, and SAP claims it's simple matter to restore. SAP hasn't released Hana support for transactional environments as yet, but CTO Vishal Sikka tells me it will certainly be adding all the high-availability and workload management tools you'd expect to run mission-critical applications. Another case of SAP confidence, but the market will have to vet these capabilities once they're available. Ramp up is expected late this year, so don't expect GA until mid to late 2013 at the earliest. --Doug Henschen
GPATEL000
50%
50%
GPATEL000,
User Rank: Apprentice
5/2/2012 | 1:19:36 PM
re: Hana and Exalytics: SAP's Hype Versus Oracle's FUD
Please do your home work and do not imagine. This is real stuff. HANA simply reads the data from disks on startup that is intitially required in RAM. No transformations are required since the data is written and stored once and transformations done dynamically as the data is accessed in main ram.

Businesses are screaming for realtime access to OTLP data. It has always been so problematic and the pain runs deep. Data latency is problematic and data duplication baloons the storage cost of traditional warehouses
Sam Iam
50%
50%
Sam Iam,
User Rank: Apprentice
5/2/2012 | 1:32:33 AM
re: Hana and Exalytics: SAP's Hype Versus Oracle's FUD
The largest issue is that DRAM is still not a persistent state storage medium. If your server crashes or you have a memory error, all of that data goes away. I imagine SAP would recommend you keep a spare copy on SSD or HDD, but how long will it take to transform the data into whatever format is required for HANA DRAM, ensure that the logs are concurrent, load a few TB of data back on to DRAM, etc? Probably quite a while. That isn't going to work in an OLTP order management system or a similar workload with constant IO. They will need a RAC or PureScale equivalent, some technology that can keep the HANA DB up and running with a primary server failure, before it takes on Oracle or DB2.

As you mention, I don't think there are a bunch of people screaming about the need for millisecond response time from OLTP workloads, especially in SAP's core industrial and retail accounts. In your average tire manufacturer or food CPG company, there are no "transformational" effects of getting .01 instead of .24 second responses to queries. It is kind of a solution in search of a problem. In OLAP workloads, in-memory BW has been around for awhile, such as IBM Cognos TM1 or Oracle Hyperion Essbase. There may be specific applications where this makes sense, but not across the board until DRAM gets to be on par with the cost of HDD.
Register for InformationWeek Newsletters
White Papers
Current Issue
InformationWeek Tech Digest, Nov. 10, 2014
Just 30% of respondents to our new survey say their companies are very or extremely effective at identifying critical data and analyzing it to make decisions, down from 42% in 2013. What gives?
Video
Slideshows
Twitter Feed
InformationWeek Radio
Archived InformationWeek Radio
Join us for a roundup of the top stories on InformationWeek.com for the week of November 16, 2014.
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.