Exadata X3 – Sound The Trumpets

It’s crazy time in the world of Oracle, because Oracle OpenWorld 2012 is only a week away. Which means that between now and then the world of Oracle blogging and tweeting will gradually reach fever pitch speculating on the various announcements that will be issued, products that will be launched and outrageous claims that will be made. The hype machine that is the Oracle Marketing department will be in overdrive, whilst partners and competitors clamour to get a piece of the action too. Such is life.

There was supposed to be one disappointment this year, i.e. that the much-longed-for new version of Oracle Database (12c) would not be released… we knew this because Larry told us back in June that it wouldn’t be out until December or January. Mind you, he also told us that it would’t be ported to Itanium, yet it appears that promise cannot be kept. And now it seems another of those claims back in June was incorrect, because yesterday we learnt (from Larry) that Oracle Database 12c would be released at OOW after all. How are we supposed to keep up with what’s accurate?

Also for OOW 2012 we have the prospect of new versions of the Oracle Exadata Database Machine, the Exadata X3, to replace the existing X2 models which have now been in service for two years. The new models (the X3-2 and the X3-8) don’t represent a huge change, more of an evolutionary step to keep up with current technology: Oracle partners have been told that the Westmere-based Xeon processors have been swapped for Sandy Bridge versions (see comments below), the amount of RAM has increased, the flash cards are switching from the ancient F20 models to the F40 models which have better performance characteristics as well as higher capacity (and my, don’t they look just like the LSI Nytro WarpDrive WLP4-200?)

One thing that doesn’t appear to be changing though is the disks in the storage servers, which remain the 12x 600GB high performance or 3TB high capacity spindles used in the X2-2 and X2-8. I’ve heard a lot of people suggest that Oracle might switch to using only SSDs in the storage servers, but I generally discount this idea because I am not sure it makes sense in the Exadata design. The Exadata Smart Flash Cache (i.e. the F20 / F40 cards) are there to try and handle the random I/O requests, as is the database buffer cache of course. The disks in an Exadata storage server are there to handle sequential I/O – and since all 12 of them can saturate the I/O controller there is no need to go increasing the available bandwidth with SSD… particularly if Oracle hasn’t got the technology to do SSD right (maybe they have, maybe they haven’t – I wouldn’t know… but working for a flash vendor I am aware that flash is a complicated technology and you need plenty of IP to manage it properly. My, those F40 cards really do look familiar…)

Exadata on Violin? No.

Of course what could have been really interesting is the idea of using the Violin Memory flash Memory Array as a storage server. Very much like an Exadata storage cell, the 6000 series array has intelligence in the form of its Memory Gateways, which are effectively a type of blade server built into the array. There are two in each 6000 series array and they have x86 processors, DRAM and network connectivity as you would expect. On a standard Violin Memory system you would find them running our own operating system with our vShare software, as well as the option to run Symantec Storage Foundation, but we have also used them to run other, extremely cool stuff:

Violin Memory Windows Cluster In A Box

Violin Memory OEMs VMware Virtualization Technology

Violin Memory DOES NOT run Exadata Storage Server

Ok that last one was a trap… Exadata storage software is a closed technology that can only be run on Oracle’s Exadata Database Machine. But ’twas not always thus…

Open and Closed

The original plan for Exadata storage software was that it would have an open hardware stack, rather than the proprietary Oracle-only approach that we see today. We know this from various sources including none other than the CEO of Oracle himself. It would have been possible to build Exadata systems on multiple platforms and architectures  – there was a port of iDB for HPUX under development, for example (evidence of this can be seen on page 101 of HP’s HPUX Release Notes). Given that Oracle’s success as a database company was founded on that openness and willingness to port onto multiple platforms, or to put it another way the freedom of choice, it came as a shock to many when the Sun acquisition put an end to this approach.

Now it seems that Oracle is going the other way. The Database Smart Flash Cache feature is only available on Solaris or Oracle Linux platforms. Hybrid Columnar Compression, an apparently generic feature, was only supported on Oracle Exadata systems when it was first released. Since then the list of supported storage for HCC has grown to encompass Oracle ZFS Storage Appliances and Oracle Pillar Axiom Storage Systems. Notice something these systems all have in common? The clue is in the name.

So what can we learn from this? Is Oracle using it’s advantage as the largest database vendor to make it’s less-successful hardware products more attractive? Will customers continue to see more goodies withheld unless they purchase the complete Oracle stack? Have a look at this and see what you think:

Oracle Storage – The Smarter Choice

This is a marketing feature in which Oracle explains the “Top Five Reasons Oracle Storage is a Smarter Choice than EMC“. But hold on, what’s reason number five?

So Oracle storage is “smarter” than EMC because Oracle doesn’t let you use an apparently-generic software feature on EMC? That’s an interesting view. Maybe there’s more. What about reason number four?

Oracle storage is “smarter” than EMC because Exadata software – you remember, that software which was originally going to be available on multiple systems and architectures – only runs on Oracle storage. Well duh.

Life Goes on

So here we are in the modern world. Exadata is a closed platform solution. It’s still well-designed and very good at doing the thing it was designed for (data warehousing). It’s still sold by Oracle as the strategic platform for all workloads. Oracle still claims that Exadata is a solution for OLTP and Consolidation workloads, yet we don’t see TPC-C benchmarks for it (and that criticism has become boring now anyway). Next week we will hear all about the Exadata write-back cache and how it means that Exadata X3 is now the best machine for OLTP, even though that claim was already being made about the V2 back in 2009.

I am sure the announcements at OOW will come thick and fast, with many a 200x improvement seen here or a 4000% reduction claimed there. But amid all the hype and hyperbole, why not take a minute to think about how different it all could have been?


3 Responses to Exadata X3 – Sound The Trumpets

  1. kevinclosson says:

    Great post, Flash. One small nit regarding the following quote:

    “Oracle partners have been told that the Nehalem-based Xeon processors have been swapped for Sandy Bridge versions”

    The X2-2 is Westmere-EP (Xeon 5600) based and X2-8 is E7 Xeon (a.k.a, Westemere-EX).

    The X2-2 will get a Sandy Bridge refresh but the X2-8 is stalled because it is a glueless 8S server and Sandy Bridge cannot function in s a glueless 8S design. Sandy Bridge processors only have 2 QPI links. Oracle will play this down though, I’m sure.

    I might recommend the following article as it relates to your words on Oracle comparisons of their standards-based storage to industry-leading offerings:


    • flashdba says:

      Damn Kevin, of course it was Westmere, I had a brain-fade. I’ll update the article to say Westmere not Nehalem but I have to acknowledge your correction and my mistake.

      • kevinclosson says:

        I understand the brain-fade. I had two, count ’em, two such freezes yesterday having to do with what CPUs are in the X2 storage cells. I went off track with a tweet about X5670 and was reminded by @acolvin that they are L5640 and then I miss focused on 5640 as opposed to “L”5640…only 35w power draw difference 😦 Ugh.

        Nonetheless. X3 is staying with 15K RPM 600GB SAS and a single active 40 QDR IB. Combined scan rate (HDD + FLASH) however increases from 75GB/s to 100GB/s. That means when someone upgrades to X3 their queries become more selective? Critical Thinking explains what that asymmetrical architecture problem is a problem. Here we see constant bandwidth but increased supply-side (product to consumer). That is no way to build a system.


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