Should we consider EXADATA?

[Hanry Zhou]

Currently we have Oracle RAC & DW running on HP itanium servers and XP storage. Now we are in the process of considering to scale up the environment. One of option we are thinking is EXADATA.

Nobody here ever worked on EXADATA, and has no knowledge about ti,  but feel this is quite a lot of different from traditional Oracle + UNIX servers architecture. It is said a appliance for Oracle OLTP application, the performance should get improved a lot. We are not a big company, and this Oracle DW is only major application running on UNIX environment, and none of us here knows EXXDATA.

Is EXADATA should be the direction we should be gong towards? 

Should we still continue on the format of running Oracle on UNIX servers rather than this "Oracle appliance"?

What are pros and cons to implement EXADATA?

Thank you very much for you inputs!

[Duncan Edmonstone]

Well my first response would be why move off HP Integrity? You can just as easily implement RAC on HP-UX as on any other platform, and unless you think you'll need to quickly move to Oracle 12 when it comes out, the HP-UX platform is good til 2018 regardless of all the nonsense with Oracle. If you think you really need the "killer" IO performance that Oracle bang on about with Exadata, you will probably find you can do better with a flash memory array such as this:

http://www.hp.com/go/vma

The cost of that will be significantly less than Exadata.

If you are determined to get off HP-UX, then the first place I'd look would be at the Proliant DL980 - on this you can run standard releases of Linux, and it also teams up really well with the VMA I mentioned above.

The issues with Exadata are:

- Horrendously expensive

- Completely proprietary - once you are on it, it will be very hard to get off (do you trust Oracle not to jack up their support prices? Or release a new one that needs a forklift upgrade with all the associated costs)

- It's an appliance - so how do you integrate it into your IT environment? How do you monitor it? How do you back it up? Yes there are ways to do all this, but chances are not the way you currently manage your environment.

Generally Exadata works for Oracle into large accounts with lots of databases they want to consolidate onto the one appliance - I'm not sure the numbers would stack up for a company who have just one Oracle DW they wanted to put on it (unless it was very large and very CPU/IO intensive)

[Hanry Zhou]

Your points are well taken, and make sense to me.

One follow-up question:
Why is EXADATA good for consolidating large accounts onto the one appliance?

I need to fully understand this point before I bring it up to my management  Thanks!

To answer your very first question, one of main reasons that we are moving off HP integrity is because of the Oracle news. I have fought for continuing on HP platform, but the management has concluded that we are going to lose Oracle database if we stay with HP and that is unacceptable for them. Honestly, understood we are fine until 2018, but what will happen after that? With this uncertainty, and plus we are planning to expand the capacity of these servers anyway, the management wants to take this opportunity and use the budget move off HP, and then now towards EXADATA option.

[Duncan Edmonstone]

>> Why is EXADATA good for consolidating large accounts onto the one appliance?

Because its big and expensive, so it only makes any kind of sense in that scenario - personally I still wouldn't use it even then - there are still better value and more efficient ways of acheiving database consolidation - not least with the DL980 and VMA.

If your management are set on getting Exadata, then someone needs to quickly make everyone look at the costs of some other options. Honestly I've seen bake-offs where Exadata with a cost measured in millions was beaten comfortably by DL980/VMA with a cost measured in hundreds of thousands....

[Hanry Zhou]

Duncan and everybody else,

It is really confusing to us to make a decision now on which platform we should pick. With the uncertainty of hp integrity & Oracle, the next choice we will have to go to Oracle(Sun/Solaris). Once we approach Oracle, they will recommend us to pick EXADATA. What about the option of Oracle on Solaris? Would it be better than Exadata?

I hate to give up HPUX, but it seems that I don't have the ground to fight for.

Thanks very much for your inputs.

[Hein van den Heuvel]

Doesn't seem confusing to me.

Seems to me Oracle has proven three times over  to have now respect for your choice of platform.

They made is impossible create a long term solution using your existing choice.

They can only recommend from their menu for processort, they can only recommend their storage.

I would not want to deal with a partner like that or at least minimize my dealings and dependencies.

Following their suggestions, instead of 'Open' you'll end up  totally dependent on Oracle for everything with no other place to go (easily).

All of that for a premium start price, premium support price, premium consultancy fees.  Hmmm.

Go Oracle for the DB to minimize the software developen cost, but pick your own alternatives for CPU, OS  and storage. probably intel/AMD, probably Linux, and any of many excellent storage choices - big or small.

fwiw,

Hein

[Alzhy]

Well for one, Oracle recently announced to no longer support HP-UX/Itanium for their future endeavours (i.e. Oracle RDBMS 12c and beyond as well as the other enterprise stacks).  Oracle actually was the last of the big enterprise software makers to abandon the Itanium ship -- earlier Red Hat (Linux) and Microsoft (Windows and SQL Server, etc) anounced their future products will no longer be dveloped for Itanium. (RHEL 6.x and WIndows 8 will not be available on the Itanium).

As for an EXADATA solution - well - if you're a very large enterprise who demand no hiccup from your Oracle RDBMS -- then it could be a viable solution considering it is an integrated super stack of hardware, storage and software. But it really all depends on your bean counting, cost benefit analysis and overall needs. It may also depend on your well entrenched Database fiefdom (if you have one.)  One caveat -- you will be beholden (forever) to Oracle.  But it shouild be noted ex-HP CEO Hurd seem to have been successful leading sales efforts for Exadata.

Note that EXADATA basically is a Linux Stack from what I've read/heard. Storage is likely Sun ZFS/Comstar technology with Infiniband interconnects to the Linux CLuster. So if you or someone in your enterprise knows what they're doing -- YOU can BUILD an Exadata like Oracle Database Machine. The components are available -- Oracle Linux, Infiniband adapters that are fully supported and tweaked in the Oracle Linux UEK OS plus Oracle Solaris ZFS/Comstar as the Storage Backend on Infiniband interconnects.

[Hein van den Heuvel]

You mention DW as Data Warehousing right?

That's where Exedata focussed on in V1.

And it is where it can help extra, notably when the optimizer can push table (partition) scans ans basic filterring into the storage engine... at a price... $$$

With V2 there are more provisions for OLTP and consolidation applications.

How much storage, and compression is roughly needed.

You probably need to need 10 or even 50 Terabytes of real data to be able to justify the price to get going.

I'm just back from the HotSOS 2012 Oracle Symposium where some interesting session were devoted to Exadata by companies like Pythian (Mark Fielding) and Enkitec.

You may want to check out  Alex Gorbachev’s webinar "is exadata for your" at:

http://www.pythian.com/library/webinars

And Kerry Osborne had some good stuff:

http://www.hotsos.com/sym12/sym_speakers_osborne.html

Good luck,

Hein