2012/11/30

Exadata "marketing" and reality

Sorry for the marked absence of posts, folks.  This year we've been upgrading all our Oracle dbs to 11.2.0.3 and all our MSSQL dbs to 2008R2 - hectic is an understatement for how things have been!

On top of that we've been told we may need to increase the size of our main Oracle dbs by 10X in the next 2 years. So, in the process of upgrading I had to ensure the groundwork for that sort of expansion was in place.  Not easy - believe me - when one doesn't even know what the hardware platform is going to be for that expansion!

All we can do is make sure whatever that ends up being, the database setup and structure will be such that the future hardware will be able to respond to the increased workload, without any artificially imposed database-related restrictions.

I'm thinking small detail things.  Like for example: ensuring all relevant tablespaces are using ASSM and bigfiles.  And so many other little doo-dahs without which, no matter what the hardware, we'd hit the wall if we didn't do it upfront.

Of course we are now going through the pains of selecting a future platform.  Which might be "IaaS"!...

For the Oracle space, we are starting from a highly evolved IBM Aix 7.1 P6 elastic private cloud platform that contains not only our database servers but also *all* our Oracle-related application servers as well as all the SOA-related midleware, with a few exceptions covered in Linux and/or VMware.

And of course: in the last few months we have been subjected to intense bombardment from what Oracle calls Exadata "marketing".  Almost constantly...

In the last week or so, I've seen even more incredible claims being made for the Exadata platform.Go read that dross now, please.  But come back here after you finished: I'm not done with it yet. Far from it!

I've already commented in a couple of places on its totally wrong conclusions.  But in all fairness to the owners of those blogs: I should also comment here rather than impose my views on theirs!
Sorry, folks. Consider this my apology for not having had the time to do it earlier!

Let me just say this upfront: I do NOT HAVE ANYTHING AGAINST Exadata - the product!  There: is that sufficiently CLEAR?

Exadata is a great piece of gear, eminently suitable to the task it was designed for.  Which is to be a database machine.

If it is the best or most cost effective for such is debatable and I am open to other opinions.  But without a doubt it is one of the best database machines ever made available.

And for those who have followed this blog for years, you can see how much I like the design: I made the "No moore" series of posts in 2006 and early 2007, describing portions of its architecture LONG before Exadata existed as a product!

As such, I have an axe to grind when I see its value being totally mis-represented in the above piece of imbecile marketing from Oracle!

I don't know who came up with the idea for that TCO "comparison". But let me just say something very clear about it:

its total and undeniable IDIOTIC approach to "comparison", if anything, will hurt Exadata more than favour it!

At least amidst folks who haven't yet outsourced their entire thought process!

Why?

Well, let me just say something that is VERY obvious:  would you do a TCO comparison between a F1 racing vehicle and a family SUV and then claim that because the F1 vehicle can run at 300mph the SUV is of little value to go shopping on weekends and ferry the kids to-and-from school?
 
No?  How DARE you!
Why not?

Let me hazard a guess: because such a comparison would be totally and completely imbecile?

Narh!, of course not!  :)

You see: Exadata is a database machine.  Like the Britton-Lee boxes of the 70s/80s.  Which later were absorbed by Teradata while their OS and software became Sybase.

As such, it is a very highly SPECIALIZED piece of machinery.  Made to do one task - and one task only - very, very well and very, very fast!

No doubt or qualm whatsoever about that from me.  Far from it, in fact!

What it is NOT  by ANY stretch of the imagination - other than what passes for the mind of the imbecile who created the above marketing piece - is this: a general purpose computing platform!

And if one buys a Exadata box in the hope of using it to run a series of applications against Oracle databases, they may end up having a slight problem trying to run such applications inside the Exadata hardware!

Mostly because it cannot run ANYTHING other than Oracle databases!

As such, to come up with a TCO comparison that CLEARLY and obviously ommits the ADDITIONAL cost of purchasing the NECESSARY application running hardware to match and drive the Exadata database machine, is my definition of complete lah-lah-land.

And a great dis-service to the Exadata folks and those who have invested time in learning its intrincate internals!

You see: the IBM P7 hardware is what normal folks - not lunatic and totally IT-ignorant marketeers - call general purpose computers.

In simple terms: you can run just about ANY piece of application AND database software in them.
Same box, virtualized, in just about any quantity and mix!  In fact, they actually can do more: they can even run DB2, Oracle, Peoplesoft, SAP, Confusion, whatever, ALL at the same time and concurrently, in the same box.

Heck: our P6 - the previous generation - runs AS400 emulation with DB2, Oracle 10g for Peoplesoft AND Peoplesoft application servers, a 11gr2 DW, a general purpose 11gr2 database for about 10 other applications, Apex web servers - AND a large mix of application servers,  ALL in the same box, without missing a beat.  Our uptime before we re-loaded it with 11gr2 and Aix 7.1 was in excess of 3 years.  I don't even know what that is, in days! But I'm willing to bet around 1000?

In simple terms again: general purpose computers are a SUV that can be used for a great many tasks.

Hence their intrinsic value: one doesn't have to buy ANY OTHER processing platform!

You buy an IBM P7, you don't need to buy another computer: it's all there to run ALL your general data centre needs, from data processing - to number-crunching - to output preparation.  Virtualized or not for quick provisioning, is entirely your CHOICE.

OF COURSE, you need to buy storage!  That is NOT the point! We are talking computers, not disk farms!

Try that with Exadata?  Slightly difficult, isn't it?  It only runs Oracle databases- not DB2, not ANYTHING else.

What's that you saying? Exadata has all the storage needed?

:)     REALLY???  All of it?

Where then do you store the interface data files and data sources/sinks for your databases? And the software for your applications?  And all the rest needed to run a COMPLETE data centre - not just one piece of it?

To do anything with Exadata, one MUST spend ADDITIONAL money purchasing the hardware to do the rest!

Now, mr Imbecile Marketeer: WHY was that ADDITIONAL cost NOT taken into consideration in that idiotic TCO comparison?


Does it even register with this kind of person the dis-service they have just given to Exadata and the folks who invested their time on it?

Because ANY "decision maker" with one picogram of intelligence will CLEARLY see the imbecile nature of that comparison and how useless it really is.

And very likely promptly dismiss any thoughts of purchasing such hardware from a company that promotes this sort of idiocy!

Sorry, but it needed to be said.  I'm sick and tired of these idiotic "campaigns" designed to seduce IT-ignorants based on "Ooooh-Aaaah!" instead of solid, simple, TRUE AND SENSIBLE FACTS!
 And there are PLENTY of such to support Exadata without resorting to imbecile arguments!

Enough is enough!

And that's about it for the time being.

I've been somewhat active this year with the SydneyOracle Meetup folks and we still have one more meeting before the end of the year.
It's been a lot of fun, with lots of new folks and students showing up.
And a bit challenging: we have to structure our presentations in such a way that they can be interesting for students and beginners, mid-journeyman and folks that have been at it for a LONG time! 

But we've had a lot of help and the good folks at Pythian have been very gracious in continuing their sponsoring of our meetings.  Thanks heaps, Paul, Alex, Yury and the rest of the gang!

(Hello, mr Oracle: it might be in YOUR interest to also sponsor here and there. After all it's YOUR software that we are talking about.  And we have a LOT more attendees than that nonsense you sponsor under the flag of "Australian Users Group".
Just saying...)


Or rather - see who came visit after a nearly two year absence due to a dishonest builder destroying their habitat near my house:



Yes!  The little green tree frogs that were so frequent around here are finally back!  Stuff you and your construction company very much, mr Triguboff: you're right up there with the Scott Cassins of this world, in my book...

These little fellas are between 1/3 and 1/2 inch long and are absolutely GORGEOUS to observe in their natural environment.  Live and let live.

Recently I spent some time down the South Coast, near the industrial town of Port Kembla.  It's amazing how this sort of thing can exist only a couple of miles from a major industrial complex:

Suddenly, there were people, seaguls, a pelican and a mini-horse in the frame!  All in perfect communion, all unaware of what goes on further down the estuary.  I love this place!


Anyways: catchyalata, folks!

2 Comments:

Blogger Joel Garry said...

Regarding predicting Exadata, look at the last post here "...but I don't know how many people
would go for that."

(I found that when I was looking for old usenet threads that included Niall and DDF 'cause of this: https://twitter.com/joelgarry/status/266316231794098177 Just about fell out of chair laughing when I realized the prediction, even if a bit of a stretch and the implied connection between RAC and Exadata is screwy.)

Wednesday, December 19, 2012 8:19:00 am  
Blogger Noons said...

Thanks for swinging by, Jgar.
Hmmm... Dunno I would classify a "RAC-only box" as a "prediction" of Exadata. That is indeed a far stretch.

The last thing Exadata is notable for is its in-built RAC.

Yes, yes: I know a LOT of folks have invested a LOT of time in RAC and are FURIOUS that anyone DARE say RAC is not God.

(It isn't what I said, but hey: who am I? I don't even write books or belong to the "Church of Ace" or the "University of wiki-URL". Much better to just discredit me. Particularly in forums which I will never directly frequent, like Twitter...)

But, you know me: since when has someone else's dumbness stopped me from saying the truth? While I'm there: I wonder what all who invested mega-$$$$$ and time in OAS training think now of Oracle's swing to Weblogic?
Ah yes: it must OBVIOUSLY also be my fault it happened and it was all to waste?

Anyways:

Much, much more important is the offloading of queries and their predicates to the first level of disk controller.

Which has nothing to do with RAC and all to do with query bandwidth. Plus the HCC stuff, which again is not specific to RAC (it can be switched on even without Exadata being present - sshhhhhh!).

Exadata is notable for offloading the query and predicates to hardware controller layers and retrieving only the row results.

Rather than sending all those db blocks back and forth between disk and db cache to finally compare and return what is a much smaller amount of data. I think the nomenclature Oracle uses for that is "Smart Scan".

THAT is the real IO advantage of Exadata and why it can beat the crap of all other similar systems WHEN doing FTS or larger queries.

Which is precisely the mechanism I described in my "No Moore" blog series.
In fact, I went a LOT further: I expected to see the SQL code in the disk controller itself - as in the little PCB that goes in the back of each disk nowadays. Miniaturization has now reached a sufficient level to allow that sort of stuff - have you seen the latest micro-PCs? Amazing stuff!!!

Without that "Smart Scan", Exadata would simply be yet another cobbled together db server with heaps of controllers, memory and CPU. Albeit one done properly and with the correct gear.

And yes: RAC. But that was never a guarantee of performance: its main purpose is to guarantee HA and scalability for very large numbers of small queries (IOW: lots of IOPS).

Anything else and it doesn't really provide any advantage against comparably costed systems.

But all that has nothing to do with general purpose computers. Which was the subject of this one, after all.

Wednesday, December 19, 2012 7:16:00 pm  

Post a Comment

Links to this post:

Create a Link

<< Home