RDBMS license costs and Open Source

RDBMS license costs and Open Source

Summary: When Sybase wants me to buy eight processor licenses at about $38K each for a Sun T5120, my natural reaction is to reach for MySQL instead because I'd rather spend my $300K on people than licensing.

SHARE:

A comment by frequent contributor JesperFrimann last week caused me to take a closer look at licensing practices among major commercial database vendors other than Microsoft.

All three of the big guys (Oracle, IBM, and Sybase) follow the same basic sales model:

  1. publish nominal list pricing only;
  2. obfuscate package names, functionality, discounts, and prerequisites to the point that no customer can reasonably understand what's where;
  3. frown severely on anyone who even thinks about publishing pricing, performance, and/or functional information; and,
  4. give the sales crew lots of leeway in dealing with customers who pass credit check.

Thus Oracle, for example, publishes its commercial price list and from that you can discover that licensing for Oracle's Enterprise Edition Plus Business Intelligence Suite lists at $295,000 per processor - but if you want to know what's in it or what it really costs, you'll be doing some serious reading, and lots of guessing, before eventually sitting down with a sales guy.

I don't mention that particular package here to pick on Oracle but to illustrate the impact licensing has on hardware choices:

  1. first, the ratio of hardware to software pricing used to make some rough sense: when a processor cost half a million bucks, paying $100K for a license for the software on it didn't seem unreasonable - but now paying $300K for a Sybase license to run on a $22K T2 seems absurd.
  2. and, second, if you really had to pay $295,000 per core for an Oracle BI license, you'd feel a lot of pressure to run it on the fastest single processor box you could find - even if that machine wasn't remotely competitive on its own.

Put these two things together and paying $260K for a pretty basic p570 running $600K in Oracle licensing on two cores doesn't seem so absurd - and is nearly half a million cheaper than looking for the same delivered performance on a four license AMD machine.

On the other hand this kind of license pricing is about to hit two rocks: one a new processor from Sun, and the other Open Source.

Sun's Rock processors will come with LDOM and hardware scout technology that should make a license purchased for operation on a single processor more productive than multiple licenses purchased for Power6 or Niagara gear - and that's going to leave these RDBMS companies facing a lot of pressure from customers and hardware partners alike: all trying to leverage more realistic pricing on their gear by threatening migrations to Rock.

Once customers start to think in terms of possible change, however, change to open source may seem much less emotionally charged than hardware change - because people identify with their hardware suppliers but, except for wintel bigots, typically much less so with software suppliers.

In other words, most people who start by considering the cost of saving half a million bucks on licensing by switching from IBM to Sun on hardware, will soon jump to considering keeping their IBM stuff but saving all of the licensing cost by switching to open source - and most will find that sticking with IBM while abandoning DB2 for MySQL feels much more comfortable than switching to Sun hardware.

Equally importantly, the entry barriers to change are lower too. If you're facing a $338K charge to put Sybase ASE on a T5120 and some consultant or blogger tells you that MySQl (or PostGresSQL) can do the job just about as well for free, your cost of finding out whether that's true or not is largely going to consist of some staff time because all the tools you need for the experiment are free - and once the test system demonstrates feasibility you're going to understand that the real bottom line here is simply that spending that $338K on staffing and support is a lot better than spending it on licensing.

Topics: Processors, Data Centers, Data Management, Enterprise Software, Hardware, IBM, Open Source, Oracle, Software

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.

Talkback

15 comments
Log in or register to join the discussion
  • Software Engineering

    One of the basic tenants of SE is that 90% of the cost of software "projects" (development) is in the maintenance phase. IOW all the work of designing, developing and deploying only amount to 10% of the lifecycle cost. So if you doubled your project dollars, you could have 10 separate (and concurrent) development projects competing to be chosen as the 1 project to be deployed.

    This of course gives you 10 times the chances to "get it right" the first time. With the probability that a project will fail being so high, this approach would seem to be "best".

    But other costs of having 10 development teams + hardware would make this impractical. BUT would it make sense to add 10% to the cost of a project to have TWO concurrent teams developing? That doubles your chances of success at a pittance of cost.

    So what I am trying to say is this: You should always have both a FOSS and a "traditional" development team working on every project. When that last "gate" comes up before you deploy - you can choose which one to go with. I call this the "Put your money where your mouth is" methodology. If Windoze or Noracle fulfills the requirements best - go with that. Pretty simple.
    Roger Ramjet
    • Agreed

      I fairly oftgen suggest the companies who think development required pay for two parallel projects and then pick a winner.

      It's expensive, most people fight tooth and nail to avoid doing it, but it works.
      murph_z
      • Only expensive in the short term

        I have witnessed the proof that 90% of the cost is in the maintenance phase. Ford paying IBM to support the (long) dead AIX 5.1 is a cost that was not amortized when the project was launched. Paying for the last 2 P690's (Power4) servers off the assembly line (@ $2 million) - to expand a project that was afraid to try AIX5.3 - added to the maintenance phase total.

        Having a few FOSS developers around with some DEV servers doesn't come close to those costs.
        Roger Ramjet
  • RE: RDBMS license costs and Open Source

    I think users should seriously consider Firebird as well.
    pbeach
    • Agreed (NT)

      :)
      murph_z
  • Beat the Drums on this one

    I wish you would return to this subject frequently--it's got
    all kind of mileage to it. The second points 1 & 2 you
    make are right on the money (pun intended). Figures such
    as $295,000 per CPU are eye opening if not unexpected.
    You could post different scenarios and their associated
    licensing costs as part of a theme. The licensing is
    ludicrous and deserves more attention than it gets.

    One thing you left out was the games vendors (Oracle)
    plays with CPU types: a Niagara I is 1.25 CPUs, a P6
    processor is X, etc. Just to show how really torqued up it
    is.

    From a technical perspective, I admire Oracle, but I'm tired
    of seeing server platform choices made on the basis of
    licensing costs and not <gasp> the actual hardware
    requirements for the application(s). (Say "application" if
    you're Wintel, say "applications", plural, if you know what
    the Fair Share Scheduler is.)

    John

    John
    jcawley
    • All week (NT)

      ;)
      in one form or another.
      murph_z
  • And, as we move to cloud computing, cloud vendors are not going to offer

    databases that they have to pay per CPU.
    DonnieBoy
    • oh the cloud...

      Why on Earth would you move your databases into the cloud?
      Erik Engbrecht
      • Perhaps...

        ...Donnieboy has an application that would benefit from infinite latency and a high packet loss rate.

        Or perhaps that's not the cloud he means. Maybe the servers are located on the S.H.I.E.L.D. helicarrier.
        dave.leigh@...
      • Agreed

        I suspect the cloud idea is most popular among home PC users with dismal personal records and no third party production responsibilities.
        murph_z
      • I wouldn't, you wouldn't

        But would Mike Cox, his pals at the restaurant, and all their bosses?

        This blog over a period has demonstrated that several contributors, and not least PM himself, consider that IT decisions are not always taken in the optimum way by the optimum types of people.

        If $name of widely sold hardware or software or IT service you most hate and despise$ was bought by many ...

        QED
        Ross44
  • RE: RDBMS license costs and Open Source

    I truly believe that software prices have really rocketed! I mean... you take the development and marketing costs and deduct them from the unit price times unit sold and what you're left with is a profit that would make a Columbian drug lord green with envy!
    Now really! Think about it! Have software companies really lost it?!
    Kostaghus
  • RE: RDBMS license costs and Open Source

    I agree the Oracle prices for the T5XXX niagara based servers is.. well.. not right. I don't know what Mr. Ponytail did to p*** off Larry but he'd better start sending flowers, and make up.

    But if you look at for example your great hate Object Evil Business Machines, then their pricing structure with PVU's is really ok IMHO, Ok for Niagara. And I could imagine the good old DB2 triple E would run just very well on a Niagara box.

    And with regards to rock... well.. late, later, rock.

    // Jesper
    JesperFrimann
  • SQL Server anytime!

    We did some performance tests after converting bits of a Sybase database (which is very similar anyway to SQL Server) and SQL Server was more than 2 times faster running it on my PC (well spec) compared to a 6GB unix box. We run simple tests. and complex end of day batches. Wait for this, my PC was faster in a very complex query than the PROD box (same number of rows for the tables), and I have no idea what the spec is but can hear the guys talking about 20-30 virtual cpus on the prod boxes. Embarassing really for Sybase, I can't wait to move to SQL Server and save a fortune but will take a while, more than 2000 stored procs. Pain to convert C unix code to talk to SQL Serve too.
    panoslondon1