SAP HANA and the death of the mainframe

Only ten years ago if you wanted to store a lot of data, you went and talked to one of a small number of vendors that sold equipment that would support it. You paid a handsome sume of money to either Digital (Compaq), HP, IBM, Sun or perhaps Silicon Graphics, Fujitsu and Sequent, and they gave you a massive computer which you hooked into a power station.

Then you went to talk to either Oracle, Informix or IBM and they sold you some expensive database software so you could make it work. A big database license on one of these systems can cost in excess of $10m.

The noughties: the decade of consolidation & commoditisation

During the first decade of the 21st century, this market consolidated. HP bought Compaq and with it the Digital UNIX systems, which it subsequently butchered. SGI had bought MIPS and stopped making CPUs, so HP made CPUs based on the Intel Itanium platform. Oracle bought Sun. SGI stopped making expensive custom UNIX hardware – they makebig systems based on Intel x86 and Fujitsu make supercomputers based on their SPARC architecture they share with Oracle. IBM bought Informix and Sequent.

If that confused you, don’t worry. You can still buy a mainframe from Oracle (Sun), HP and IBM. And database software from IBM or Oracle. It’s worth bearing in mind that HP’s platform is crap and the lack of a sensible roadmap (and lack of Oracle support) suggests that it is being sunsetted. To be honest, Itanium (dubbed Itanic) was a lame duck that never got off the ground. Sure, HP is suing Oracle over this, but that’s just a charade to make customers feel better.

The rise of Wintel/Linux/Virtualization dominance

Three additional dimensions have emerged. First, Intel’s x86 range of CPUs have become much much faster, and for most purposes, a single pizza box system can serve 99% of computing needs – even for a multi-terabyte database.

Second, Linux has emerged as a stable and more-or-less free (you pay for support from one of the large vendors like RedHat or SUSE in many Enterprise scenarios) Operating System. It runs on just about anything from  your cellphone to very large systems.

Third, we have achieved massive consolidation of equipment using virtualization software like VMware. This is because most systems sat there idle, so you can take 50 or 100 systems and put them on one physical piece of equipment. This isn’t that relevant to this discussion because we’re talking about large scale systems, which VMware doesn’t help.

What’s more for commodity applications, Microsoft’s Windows and SQL Server database are pretty attractive. They’re cheap and easy to use for mid-size data volumes, which isn’t really the focus of this article. Microsoft would argue that they compete with DB2 and Oracle, but this isn’t the case for databases >10TB: IBM and Oracle still rule the roost.

The stop-gap: Teradata and Exadata

First Teradata and now Oracle’s Exadata have now built custom hardware based around largely commodity components. They are both based on the Intel x86 platform and Linux Operating System and both built on largely the same premise: make everything parallel.

But Teradata and Exadata only benefit customers from a performance perspective. Despite using commodity components, they are extremely expensive and profitable for Teradata and Oracle respectively. Plus you are tied into their platform and they will come knocking for their maintenance dollars.

For my money both these technologies represent an opportunity in the market for the vendors but it is not the end-game.

Where SAP HANA fits in

SAP HANA and other solutions like Hadoop, are truly fascinating. SAP HANA runs on high-end commodity hardware and provided you have enough memory to run your database, it doesn’t matter what you run it on from a technology standpoint.

Initial certified hardware solutions are still quite expensive – I priced up a 1TB Dell system for $75k – not including disk storage, which will probably double that number at retail price, so think $150k. But really that’s nothing compared to what an equivalent 10TB database cost 10 years ago (SAP HANA compresses 10:1 compared to databases back then).

And worryingly for Oracle, Teradata, IBM and HP, it is nothing compared to what Mainframe or Teradata/Exadata hardware costs.

Where does this leave us today?

In the short term we have IBM posting higher profits from Mainframe sales, HP in all kinds of organizational trouble and Oracle focussing on Exadata. This figures as people will keep buying mainframes to stay in support. This market won’t die out for many years, but it will start to tail off by 2015.

The decline of the mainframe

IBM at a global level don’t care because they already know their hardware business is in terminal decline. They sold off their PC business to Lenovo and their services business is growing at a sufficient rate that that it’s all good. Although I’m not sure if the SVP of Hardware at IBM feels the same way – you can bet his targets aren’t decreasing year on year. Besides IBM have the best and dominant mainframe platform, for various reasons.

Teradata should know by now that they’re screwed long-term, and HP have bigger problems, like how to stay alive. Mainframes aren’t a focus for them.

Oracle remain arrogant (plus ça change!) – and in reality may innovate fast enough for this not to be a problem for them. They have an in-memory product called TimesTen in development which will no doubt compete with SAP HANA. And Oracle has traditionally run on the kitchen sink, including hardware from most of the above vendors. I suspect they will move away from the appliance game again over a period of time.

Advertisements
This entry was posted in HANA, Technology. Bookmark the permalink.

13 Responses to SAP HANA and the death of the mainframe

  1. The comparison between mainframe & HANA is interesting for more than the aspects mentioned above. Unlike Teradata or Exadata, HANA is not just a database appliance (HW + DB). It actually brings back the 2-tier architecture (HW + DB + App Engine) to the scenario, what might enable not just a technical performance improvement but a complete redesign on the business applications which run over it, adding a potential (massive) functional performance improvement as well.

    This just emphasizes even more the scenario you’ve described: Teradata might be looking at hard times in the near future, while Oracle & SAP are the true only companies that can (currently) take advantage of this business + technical (re)merger.

    • John Appleby says:

      Like your analogy – perhaps we should add Hadoop & InnoDB into the mix too as a potential continued competitors.

      I think there will still be a place in the world for traditional DBs especially mySQL and MSSQL – particularly at a departmental level. But for large-scale ERPs and EDWs? Nope.

      • Definitely. 3-tier is like the concorde. It was the top notch plane of its time, however it was put to sleep because its price per mile was too damn expensive. Luckily, in the SW industry (unlike aeronautics) there is place for several thousand of vendors, each one providing software at a particular segment of a particular market. So they’ll not die, just focus on different market segmentations.

        And who knows. With twitter & facebook creating raw data quicker than ever, in some years even medium local vendors will want to have some analytics running over petabytes of data from their consumers.

      • As for Hadoop, I think its focus is different. It does not aim at becoming an integrated platform for business operation applications, but rather to be a pragmatic solution for application execution on massive amounts of data that technology currently can’t handle in an easy way. If your business demand it, distributed analysis for (really) big data will still be required, no matter if you’re on HANA, Exadata or on a classic relational DB.

        On the other hand, with the speed technology has been growing, in 5 to 10 years, analyzing petabytes of data will most likely be possible in a non-distributed environment (but, at that time, “big data” will mean something different, i.e. exabytes of data).

      • John Appleby says:

        Well Hadoop is more generic, but Hive provides datamart capabilities based on Hadoop which might compete with SAP HANA.

        With Moore’s law, we expect to double every 18 months. So in 10 years we expect roughly x80. We currently have 2TB systems, so expect 160TB in 10 years – nowhere near Petabytes until closer to 15 years. But that’s way too far to think ahead with today’s computing terms – I struggle to see a few years!

      • Moore’s law doesn’t take into consideration the new business models implemented (i.e. social media). It basically states each ~2 years, the number of transistors that can be delivered at unit line prices are doubled. That doesn’t mean the actual demand curve for more computational power will necessarily follow the offer curve. =)

      • I think you’re undervaluing the power of MSSQL quite a bit. While in SAP Support I was shocked at the number of MSSQL installs I saw popping up in the late 2008 to 2009 timeframe. And yes I worked with large scale BW systems (MaxAttention customers). It was my impression that with the release of SQL 2008, Microsoft has caught up with DB2/Ora in terms of large BW support and scalability.

      • John Appleby says:

        I agree, for many scenarios. I have dealt with multi-TB databases that run on MSSQL.

        However for large data sets, MSSQL has an inferior calculation engine especially for BW. All of the 20TB+ BW instances I know of run on DB2.

        In the end HANA should kill all 3 viable platforms.

  2. Jon,

    I’m still a little confused by your distinctions. As I understand it Teradata and Exadata are hardware+software machines but still rely on disk as a retrieval source. Let’s call these traditional databases on crack. They read/write data in parellel (and very fast) but don’t utilize any cleve boost in I/O performance at the memory level.

    Hadoop does also go after disk, but scales more gracefully because it is generally hardware agnostic and you can take advantage of things like virtualized servers (a la Amazon MapReduce) to spin up servers in a matter of minutes rather than relying on static machines. It’s almost infinitely scalable and requires less time to scale up. (read this -> http://hadoopblog.blogspot.com/2010/05/facebook-has-worlds-largest-hadoop.html ) So in instead of waiting for a large chunk of data to come into memory they simply call hundreds of servers to do that in a fraction of the time and then combine it together.

    The way data is actually stored in BWA/HANA and Hadoop-HDFS is key-pairs, which is much different in the way Exadata and Teradata. The data storage alone offers better performance and scalable, regardless of the utilization of in-memory technology. It’s my understanding all of the solutions are trying to eliminate the I/O bound issues that arise in all of modern computing. HANA is really only unique because I/O is reduced by the advancements that SAP has made with the Intel chip and pinning data centrally in memory (and thus eliminating the need for the CPU to wait for data to be called into memory).

    InnoDB is also another in-memory technology thats pretty interesting as well. Workday is built on it: http://www.dbms2.com/2010/08/22/workday-technology-stack/

    Correct me if I’m wrong 🙂

    Cheers,
    Mike

    • John Appleby says:

      I think we’re saying the same thing, but I skimmed over a lot of the items you went into detail on (as well as a lot of other things 🙂

      The difference with Teradata and Exadata is the appliance and commercial model – they are just traditional databases on crack, agreed. But they have good sales figures for now because they solve performance problems with regular hardware in some scenarios. They don’t change the world and that’s why I think the sales for them (as they stand) will be short lived.

      The difference with Oracle is that they have other options. I think Teradata is technologically at the end of the line – they don’t have the research power and resources that Oracle do, or TimesTen.

      Hadoop is interesting for large deployments like Facebook but I don’t think it can compete with e.g. SAP HANA for sheer number crunching ability of mid-size data sets. Hadoop is true Big Data. InnoDB also interesting.

      • There is currently no connector from NewDB (HANA) to Hadoop nor an indication that there will be one, but it might very well be the case in a year or so. Who knows.

      • The problem I see with sales in this space is that is largely based on how much buzz is created by buzz words and not actual performance numbers. It’s impossible to tell a customer their reports will increase by 10x without actually using numbers. Where SAP can win is by building business apps on the platform – which shows business value, not just some “game changing” technology. I think this is where Exadata and Teradata fail. The biggest threat is actually from Netezza and Vertica since they have industry specific business cases for business analytics.

        “Hadoop is interesting for large deployments like Facebook but I don’t think it can compete with e.g. SAP HANA for sheer number crunching ability of mid-size data sets.”

        How so? MapReduce is used for a number of web applications and crunches data very well. I don’t think in today’s age that a company developed solution will ever out perform an open-source project. (that’s disregarding the other transactions that are valuable when buying enterprise software)

  3. Pingback: AdVoice: SAP HANA: the Evolution Accelerator | What is ERP

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s