Oracle Predicts the Future

While there is nothing in the recent 10-Q filing with SEC to suggest Oracle sees any clouds on the horizon, an analyst found that a number of IT professionals plan to reduce dependence on Oracle by switching from Solaris to GNU/Linux but not to Oracle’s version. They are switching to RedHat or Suse. In the process, there is a possibility of switching to a FLOSS database. The obvious motivations are lower costs of licensing. On the other hand, Oracle database does have a reputation for performance and there are lots of dependencies in the applications of large enterprises. Switching databases would have a significant cost but the long-term benefits could be substantial.

I don’t see it. Red Hat, for instance, recommends RedHat to run Oracle database, not to replace it. Oracle’s cash cow is not immediately threatened.

“Your mission-critical CRM, ERP, ETL, or DSS deployments rely heavily on an Oracle database that needs to be reliable, available, and scalable. Historically, database customers sought out the UNIX/RISC platforms to best enable those feature. But today, x86 servers have dramatically increased in performance and availability, making them a more cost effective platform than ever for running Oracle databases. Of course, to fully maximize the performance and availability features of today’s x86 hardware, you will need an enterprise operating system like Red Hat Enterprise Linus.

This paper highlights the benefits of using Red Hat Enterprise Linux as the server platform for your Oracle database implementation. It introduces you to information, customer success stories, and reference architectures that display our scalability, availability, reliability, and manageability.” see Red Hat Enterprise Linux: The Ideal Platform for Running Your Oracle Database

About Robert Pogson

I am a retired teacher in Canada. I taught in the subject areas where I have worked for almost forty years: maths, physics, chemistry and computers. I love hunting, fishing, picking berries and mushrooms, too.
This entry was posted in technology. Bookmark the permalink.

12 Responses to Oracle Predicts the Future

  1. oiaohm says:

    This is incorrect Dr Loser
    “Linux doesn’t do vertical scaling very naturally”

    Vertical scaling is very distribution linked. scientific linux a redhat relation scales easily. Runs Oracle like a dream.

    Many-cpu systems (vertical scalability) is something cern and other parties behind scientific Linux care about. They do have extra software just to manage that. Redhat changes extra for the same applications. Yes the high availability addon pack contains them.

    There is a critical factor. Do not use a standard ubuntu even if you could convince it.

    Solaris trust extensions and Linux selinux map 1 to 1. So secuirty difference between correctly configured redhat or scientific Linux to a Solaris is in fact nothing. Because they are both selinux distrobutions.

    Right Linux with Oracle no fowl.

    ZFS is why you choose Solaris if it suites you work load better. Redhat and scientific have a broader range of cluster filesystem that can be used.

    Phenom
    “memory-pinning, async IO operations, and better scheduler for many-core / many-cpu”
    Most of that tech that is in AIX has been transferred by IBM staff to the Linux kernel for x86. This is what SCO was suing IBM over.

    http://www.phoronix.com/scan.php?page=article&item=intel_3960x_scaling&num=3
    But there is a big problem with solaris.

    You can find your self application staved and even in cases like with that graphicmagic there where it does not perform on solaris at all.

    Notice how many of those testsuites don’t run on solaris.

    Yes when you are talking vertical scale you are talking IBM bricks.
    http://cateee.net/lkddb/web-lkddb/NR_CPUS.html
    With standard linux that takes to over 8000 CPU’s.

    There are a few custom Linux systems out there that take vertical to over 200,000 cpu cores. Yes that is a x86 system.

    Linux kernel max vertical is very high when you get into the custom hardware. Stock x86 designs without motherboard to motherboard numa controllers stop at a poor 255 cpu cores ie your general SMP.

    There is no known limit that Linux has on numa hardware that does not have a numa limit.

    Its the numa controllers that limit how far vertical Linux can go these days. Same thing effects solaris.

    Support of 4096 CPU was added mainline Linux kernel 2009. But that is not the limit third party patches for particular hardware exceeds this by a long way. Basically when you apply the patch for the numa controller that goes past 4096 it will alter the max.

    Basically is only that the numa controller drivers are not merged mainline why we are looking at the 4096 from mainline. So the machine will at least part work out the box. Rebuilding a kernel with 4096 cpus does not take long.

    PPC numa controller is merged mainline so on that you can go up to 8000+. Without any custom patching.

    The correct complain today is Linux cgroups are not quite as good Solaris zones at breaking down a huge vertical system into pieces. Not far behind. The fact you can use non oracle hardware that is cheap can offset the pain.

    Yes both go to the same vertical point is the management bit when you get there also the price of the hardware.

    As long as you don’t go above 4096 on x86 cpus per block you don’t have to custom build kernel. This is when a maker provided hyper-visor can be handy.

    Windows you are petty much sitting where it suxs. Limited cluster filesystem gives up the ghost at only a few hundred cpu vertical. Does not even make it to a 1000 cpus. Windows drops the ball completely at 255 cpus. MS is promising next to be able to go 512. Yar right little out of date for high performance hardware support.

    Solaris and Linux can pretty much fight it out vertical these days. Windows down right forgot it you knife taken to gun fight.

    A few years back Dr Loser your vertical claim was valid.

    Solaris and Linux come down to a workload tests.

    Solaris is just as much of a prick to manage at times as the Linux distributions particularly with applications that don’t perform.

    Oracle database is never just the database in most cases.

    The big issue with windows is limits. Linux and Solaris limits on number of openfiles ammount of ram that can be allocated.

    Just to give you a idea how big the difference is.

    1024GB-8589934592GB That is a Linux systems 2 max memory. 1024GB is a SMP 8589934592GB this is NUMA.
    128GB is Windows.

    How much ram can a process allocate if you let it go nuts. I don’t recommend this. 8388608 TB is Linux 8 TB is Windows per process memory allocation. I would not like to bench the Linux limit. Coping 8388608 TB from storage would take a huge long time. Basically need something numa to generate it.

    32bit PAE hardware also reads bad for windows. Linux 64G per process Windows 2g-3.5G per process.

    This huge different limits crosses number of threads number of files open number of connections.

    Internally Windows is crippled.

    Yes Linux desktop is less crippled than Windows is for doing high end work.

    The fact that Windows is crippled is why you don’t stand a chance converting me to Windows. OS X maybe because it less crippled.

  2. Dr Loser wrote, “A single reference, please.”

    Go to M$’s pricelists and do the maths: A $2000 piece of server can have a server licence ~$1000 and hundreds of CALs at ~$40, that comes to many times the price of the server. Of course you can spend more on your hardware to make M$ look good, but most just want stuff that works.

  3. Dr Loser says:

    @Phenom:

    Yeah, I’d go with that.

    Unlike Robert, we live in the real world.

  4. Dr Loser says:

    @Robert:

    “In a proper setup run by that other OS, the cost of the software licences can easily exceed the price of the hardware.”

    I hate to call you out on this stuff.

    However, on a cursory examination, it’s bum-fluff.

    A single reference, please.

  5. Phenom says:

    @Dr Loser,

    About 6-7 years ago I was managing the “front-end” part of a project, which used Oracle as a db. I recall some discussions with the db admins out there, who shared some articles and techniques how Solaris and HP/UX were just great for Oracle. AIX was also praised, even though not thanks to IBM. I think it was the support of things like memory-pinning, async IO operations, and better scheduler for many-core / many-cpu systems (vertical scalability).

    Probably things have changed, not sure about that. However, most Oracle shops I have contact with still clutch tightly to their HP/UX or Solaris.

  6. Dr Loser says:

    @Robert:

    I shouldn’t really have to explain this to anybody who has dealt with databases, but:

    “It costs me nothing to admininister a dozen databases on my Beast.”

    Bully for you and your dozen databases.

    Now, and genuinely, and I would like to be a part of this because I am knowledgable and greedy:

    Who do we sell this excellent concept to?

  7. Dr Loser says:

    @Robert:

    Ten days, Pog, ten days. Pick a measure: profit or turnover.

    Either way, <1.5% and you win. I will give #20 to Mellisa's anti-malarial foundation.

    The one, tiny, little thing that I ask of you: if M$ fails, either on the 1.5% profit or the 1.5% turnover basis …

    … is it possible, at all possible, that your dreams over-reached reality? If so, I'm looking forward to you admitting the fact.

    (Otherwise, to be perfectly honest, I don't give a shit. I'm just a programmer. I don't care what on. I have no agenda.)

  8. Dr Loser wrote, “Ultimately it doesn’t come down to the choice of OS; it comes down to the cost of hardware, the culture of the company, and most importantly the pool of experienced dbadmins out there. I’m not sure I’d trust a Linux one.”

    Not even close. In a proper setup run by that other OS, the cost of the software licences can easily exceed the price of the hardware. That may not be so on some monsters but one can easily find a nice $2000 server which would cost $5000 more for the licences. e.g. ~$1K for the licence and ~$40 per seat for the CALs. 100 seats on the LAN costs you $5K for the privilege of using $2K of hardware.

    Then you pay Oracle how much?

    MySQL or PostgreSQL seems so much more reasonable on Debian GNU/Linux, $0. It costs me nothing to admininister a dozen databases on my Beast. Why should it cost much at all on a more expensive system unless you just want to throw your employer’s money to the wind.

  9. Dr Loser says:

    @Phenom:

    Leaving aside oiaohm’s customary brusqueness (even possibly offensiveness) he may actually have some sort of a point here; and it’s precisely because the stats are next to impossible to get at. (Unless someone wants to leak a Gartner document, perhaps: funny how Assange doesn’t seem to be interested in that sort of “freedom.”)

    The sketchy information I have been able to trawl (and it is sketchy, and it stretches over several years and several versions of Oracle, and it is invariably volunteered by somebody with an axe to grind) suggests that Oracle DBMS throughput is about 30% better on *nix than on Windows. Curiously this appears to be precisely because Windows is a general-purpose system, whereas *nixes are in general admit of a gazillion kernel and FS parameterizations for fine-tuning a database. Of course, whether it is worth paying a Solaris or Red Hat consultant the obscene amount of money to do this is a different question. (Whilst at Visa I was privileged to see a Sun consultant with her “little black book” of these tricks. She was extremely pretty, but not $50,000 pretty, as it were.)

    There’s a division of IBM that supports Oracle on AIX, but I suspect there’s a sales culture that pushes DB2 (I forget its modern name). Last time I looked (anecdote alert! anecdote alert!) HP tended to sell on Those Other Databases (Sybase for finance, Informix — now IBM, of course — for legacy, Postgres for scientific).

    It’s just my impression, but you have three basic choices for an Oracle platform:

    (1) Windows, which is easier to administer and doesn’t require esoteric knowledge of the whole rest of the system. I’ve seen it working on a scaled-down version of an Enterprise product, and it’s probably a good choice for an SME with limited sysadmin and dbadmin resources: ie the two are one and the same.

    (2) Red Hat (almost certainly Red Hat and no other Linux, other than the Oracle offering), which allows you to tweak to your heart’s content. Good for bit-heads like oiaohm. Questionable for a large, professional, IT department.

    (3) Solaris, which comes with all the (presumed) benefits of Linux and several rather more important ones. If you have a humongous data processing operation, then Solaris is an obvious choice, if only for the vertical scaling. (Linux doesn’t do vertical scaling very naturally, and if it did you’d be looking at the IBM virtualisation solutions.)

    Whereas Robert seems to think that one size fits all, I would strongly disagree: any one of these three options makes sense, in the correct circumstances. Ultimately it doesn’t come down to the choice of OS; it comes down to the cost of hardware, the culture of the company, and most importantly the pool of experienced dbadmins out there. I’m not sure I’d trust a Linux one.

    And the only real advantage that oiaohm proposes is (if I follow him) throughput. Honestly, 30% is not that big a deal. I’d take ease of use, reliability, security — excellent on Solaris and on Windows servers, far less so on Linux — scalability, and the human costs into account first.

    It’s a demented bit-headed proposition, though, isn’t it? Here you are with a colossal piece of software, built up over twenty years, and some divot comes along and suggests that kernel patches are the way to go … I think no, oiaohm.

  10. Phenom says:

    Ohio, I dare you to back up your statements with some official documents or articles on the topic. Until then, you are mere noise.

  11. oiaohm says:

    Uninformed as normal Phenom. There is a reason why Oracle database performs so good on Linux kernels compare to the rest of the Unix world and Microsoft offering. For over the last 13 years Oracle has sending patches to the kernel to fix performance flaws that effect Oracle.

    Hello HP/UX AIX Solaris had not been receiving those patches.

    Solaris has increased in performance running Oracle since Oracle took over reason its now getting patches from the Oracle database development team. HP/UX and AIX are basically not in the game for this particular case.

  12. Phenom says:

    Actually, Pogs, the ideal platforms to run Oracle would be Solaris, HP/UX and AIX.

    Linux and Windows are not in that category, though Linux can be tweaked to some improvement over standard distribution.

Leave a Reply