Oracle Does Hardware

Well, it's been discussed for a while, but the news is out in the open: Oracle will be selling a data warehousing "appliance" (server and storage) built by HP.

In typical understated Oracle fashion, they bill this as "The World's Fastest Database Machine".

But, all posturing aside, it's an interesting example of the different trends going through the industry right now.

The Basics

If you take a closer look at the hardware, it's pretty cool. 

It looks like "storage nodes" (a server and DAS), lashed up to a larger database engine using Infiniband.  The idea is that Oracle breaks a large query into smaller chunks, pushes them down to the individual nodes, and the results are aggregated at a higher level.

Performance scaling is achieved by adding more storage nodes, and (presumably) breaking queries into progressively more equal-sized pieces.

Of course, there's little in the way of performance comparisons to help us evaluate just how fast this beast might go, except the "Up To 10x Faster" which smells a bit optimistic, never mind that it's Oracle comparing with itself, rather than other data warehousing appliances.

Interesting Bits

From a pure storage perspective, not much to talk about here other than their use of disk mirroring throughout, if I'm reading the data sheet correctly. 

Ummm .. that can get really expensive folks, especially in larger configurations.  We're not only talking disks and power here, we're talking nice HP server-based storage nodes to put them in, which don't look especially cheap to me.  If it were me, I'd want a RAID 5 (or 6) option ...

I think the plan is that Oracle sales reps will sell this stuff, and HP will do the installation, support etc.  That's certainly an interesting model.  It's not one that I'd attempt, but -- hey -- maybe they'll find a way to make it work.

I still have questions about backup, remote replication, etc. since many of these DWs are becoming business-critical systems, but I'm sure there are answers for all of this.

So, Here's What's Really Happening

As I understand it, there's a consensus in the industry that Oracle is under serious pressure from purpose-built tools in the DW/BI marketplace.

Whether it's traditional players like Teradata and Netezza, or newer players like DATAllegro (acquired by Microsoft), Vertica or ParAccell -- these specialized platforms can come in and make Oracle look very bad indeed from a pure price/performance perspective.

I think Oracle had to do something about it, and do it quick. 

One option would be to acquire and/or build software optimized for DW/BI.  From what I've heard, these specialized players can do some amazing things. 

Another option would be to work with a server vendor to build some Really Fast Hardware to run a version of their existing code.

Without hard comparisons, though, I have to say I'm skeptical.  First, Oracle has always been able to break queries into smaller pieces, push them down to other servers, and aggregate the results. 

Nothing really new here.

Second, array-based storage technology is not the bottleneck; our work with Oracle and other DW/BI environments routinely shows that we can feed data to a server just as fast as it can take it. 

"Queries run closer to the data" ???  Sounds like some creative marketing here. From a purely technical perspective, there's nothing really new here either.

Maybe they picked up some benefit from the server-to-server connection being Infiniband rather than something like RDMA over ethernet, but I'm dubious about that as well.  You'd have to show me that the ~200MB/sec bandwidth  offered by a pair of GigE connections was slowing things down.  Remember, this traffic is ostensibly reduced query results, not raw data.

Now, to be fair, you can pick up some serious performance benefits by hitting the "sweet spot" of balancing CPU and I/O -- something we've been doing with Oracle and Dell for a while, using the CLARiiON product family, but I don't think the claim is "10x" or similar.

And Then There's The Anti-Appliance Argument

There seems to be a line of tension between business users of DW/BI, and the data center infrastructure people over the whole question of appliances. 

Of course, the DW/BI users would like something optimized for the job at hand.  But the data center infrastructure people try to limit diversity as much as possible: consistent choices for servers, operating systems, network, storage, backup, replication, etc.

If this behemoth is going into an existing HP/Oracle shop, that's one thing.  But if it's going into a shop that's trying to standardize on other choices, well -- that's another thing.

And, Of Course, I'm A Bit Cynical ...

Every year at Oracle Open World, we hear about many "new initiatives" from Oracle.  Well, not to be harsh here, but it's my impression that very few of them get talked about at next year's Oracle Open World.  I routinely dig up past announcements from previous years, and it's relatively consistent pattern.  I think it's fair to ask the question -- just how serious is Oracle about all of this?

And, if they really want to solve a performance problem, we've got the answer .... :-)

Seriously, though, I've seen more than a few environments where the performance problem wasn't so much getting the data off of disk, it was the I/O storm generated by the subsequent data reduction and analysis.  And it looks like a small amount of EFD might be very helpful in some cases.

I think we'll have to wait a while to see if Oracle (and HP) is going to be successful with any of these. 

Time will tell.

[Update on Oct 20 -- I've been responded to! Go see the latest here]

Courteous comments welcome as always ...

Comments

[Alex McDonald]

Perceptive. One additional comment; this doesn't make Oracle a hardware company. Oracle is about Oracle. Thar's serious loot in them Orakule licences.

[Chuck Hollis]

Oh yes, no surprise it's all about those software licences :-)

Interesting note, did you see the results we got by virtualizing the much cheaper Oracle "SE" licenses rather than the full-boat RAC licenses, with VMware providing many of the advanced features found in RAC, e.g. load balancing, clustered failover, etc.?

There are many different ways to play the commoditizing game ...

[Bit Counter]

Chuck:
Good post, and a bit adjustment from my end. Firstly, this is not an appliance. It is a DB machine. It will need customization and maintenance - so service dollars for HP/Oracle. So even if this is sold in an HP shop, it may not hit the ground running.
Second, DMX-4 with the SSDs may have issues for a DWH type workload, esp with the
data movement of larger tables across storage tiers. Isn't this a more of an OLTP type device?

BTW, this year it was NTAP and EMC on the cross-wire during the keynote. Last year it was VMW and prior to it was BEAS. Let us just get used to it. :)

Truly anon,
Bit Counter

[Chuck Hollis]

Bit Counter

The 'appliance" word is theirs, not mine. I agree with you, this is not a simplistic device.

As far as EFDs making sense in various DW/BI scenarios, I outlined one that we're investigating (e.g. the temp space where you do your joins, data reduction, whatever) which we believe drives an interesting mix of I/Os that EFDs can make run much faster.

Random reads and writes -- that's what they're good for. And, in a DMX, the cache soaks up all the writes, so you're left with random reads. In a CX, not so much write cache, so the profile is different.

Too soon to tell exactly how and where they'll play, or when.

Someone told me that Tom Georgens from NetApp used his speaking slot to "set the record straight" on the usable capacity debate. Amazing, that is.

And you're right, there will always be a tussle of one sort or another going on at these shows.

I'm taking bets as to whether this "initiative" makes it 12 months to the next Oracle Open World. Obviously, their "virtualization initiative" (e.g. Xen instead of VMware) introduced last year isn't making the grade, among others ...

Cheers!

[Chuck Hollis]

Hi, this is Chuck again, with some updates.

First, someone pointed out an important point that I had missed, e.g. the hardware was probably developed for HP's own Neoview DW/BI environment, see here: http://h71028.www7.hp.com/enterprise/cache/414444-0-0-225-121.html

This brings up an interesting coopetition question: if the customer is interested in large scale DW, which platform does the HP rep promote?

Secondly, since the hardware platform is presumably runs something like Linux or HP-UX, you have to wonder how fast something like DATAllegro, Vertica, Greenplum, ParAccell would run on it.

Can optimized hardware make up for software architectural issues?

Time will tell ...

[Ike Kuiper]

Hi Chuck,

I agree on your scepticism, especially thinking of the big announcements of Oracle in the recent past about for instance Oracle VM or Oracle Enterprise Linux (OEL), what have we heard about this since the announcements? Not much and certainly not a lot of market-penetration.

Regarding this appliance announcement, I have talked with several Oracle-sales and Oracle-partners at OOW last week. Some conclusion: Oracles-sales does not care about hardware and do not want to discuss this with customers. Especially the application- and middleware-side of Oracle, could not care less. They do not see a huge market for them, as there a not that many customers who will pay 1.6M USD for the licenses of this appliance. Also, although Oracle's business model is evolving to more and more channel-led business (averaging around 60% of the business now), this is direct-sales only. That means that resellers and SI's are expected NOT to promote this

[Chuck Hollis]

Hi Ike -- I tend to agree

I see this as a purely defensive play by Oracle sales -- if they're in a situation where a specialized DW platform is the competition, they can drag out this beastie to look like they have one too.

I don't see Oracle sales or partners leading with it. I don't see Oracle customers standing up and saying "I want one of those".

And I bet HP did most of the heavy lifting to bring it to market.

-- Chuck

[Nazareno]

"Closer to Disk"? Has Oracle code running inside a Disk Drive?
If don't, so ther's no changes...
Oracle "Hide" CPUs and Storages inside a Box, with the same disk interfaces (FC, SCSI, etc.)...

[Chuck Hollis]

... my point exactly, there's no magic here ...

[Ckazky]

Chuck,

While EMC and other SAN storage vendors are seeing Exadata as a competitive threat, I wonder if you are missing the real competition.

Exadata (as Larry Ellison explicitly stated in October 2007) was developed as Oracle's answer to appliances like Netezza, etc. Exadata was a reactive, defensive move by Oracle, not an offensive one. It doesn't appear that it's working out so well.

But perhaps SAN vendors should also be concerned about the Netezza's of the world. More concerned than they are about Exadata. Note that in the places Exadata is losing (EHarmony and NYSE most recently), they are not losing to SAN players, but to Greenplum and Netezza.

Oracle and EMC face a common threat from the data-appliance guys, it's a classic Clay Christensen 'innovator's dilemma', because these guys are bringing technology that is first and foremost, incompatible with 'mainstream market models for making money'. Perhaps EMC and Oracle should be working together to respond to a disruptive threat you BOTH face.

[Chuck Hollis]

Ckazky -- great point. EMC actually has done pretty well partnering with many of the players you mention, and many others.

And there are parts to the story not yet visible :-)

-- Chuck

[Justin]

Oracle seems to be saying the way to fix Oracle architecture limitations is buying more and bigger hardware.

Column based databases like Sybase IQ reduce the amount of io needed by 90% anyway, for OLAP , reporting and querying they can deliver 10,100,1000 * performance improvements on much cheap hardware.

Oracles response to new architectural dbms thinking, to build a propriertary Oracle database mainframe!