Sorting out Netezza and Oracle Exadata data warehouse appliance pricing
Netezza recently announced a new generation of data warehouse appliance called TwinFin. TwinFin’s clearest stated list price is “a little under $20,000 per terabyte of user data,” which in my opinion immediately became the new industry reference point for discussing prices in the data warehouse appliance category. Vigorous discussion ensued, especially in the comment thread to the first of the two posts linked above. Here’s some followup.
Netezza should not have claimed a “10-15X price/performance improvement,” based on a 3-5X performance improvement and a 3X decrease in price/terabyte, and I should have grilled Netezza harder when it first made the claim. In fact, there is no unit of performance that you can, in a reasonable blended average, get 10-15X more of per dollar in TwinFin than you can in the predecessor NPS series.
To look at it another way, multiplying 3-5X by 3X would only make sense if 3-5X were a measure of something like “terabytes/unit of performance.” But in fact the 3-5X is a blended average of something more like “units of performance/unit of time”; i.e., you can do 3-5X more calculations or queries in a unit of time over the same database (of the same size*) on the new machine as you can on the old.
*Since Netezza is so table-scan-oriented, processing time can be expected to be pretty sensitive to the size of the database. Specifically, to a first approximation it could be linear.
Netezza is regrettably declining to officially disclose TwinFin list prices. (Note: “Regrettably” is a euphemism.) However, if you do the math in the obvious way you indeed get to something close to the (technically NDAed) Netezza TwinFin list price. Ditto if you do the math starting with $60K/terabyte — and the same 2.25X compression figure as Netezza uses to discuss TwinFin — for the predecessor Netezza NPS series.
Note: GSA schedule prices, as linked to in a prior thread, are discounted from commercial prices.
Bence Arato estimates that the Oracle Exadata price is right around $4 million for 46 uncompressed terabytes of user data. I found Bence’s estimates excellent when he helped me work out then-current Exadata pricing last September. That’s a little under $100K/terabyte uncompressed, vs. Netezza’s figure of a little under $45K uncompressed. I would guess Oracle’s compression is a little better than Netezza’s, but only a little. I hope those Oracle figures take indexes into account (Netezza has no indexes, and the zone maps it substitutes for indexes take little space), but even if they do, there’s a considerable price difference now between Exadata and Netezza. Also, Netezza TwinFin seems to offer more processing power per terabyte of data than Oracle Exadata does — specifically via its FPGAs — giving hope it does more work as well.
Oracle’s claim of a more benign long-term pricing model than Netezza’s is (largely) specious. The Oracle argument goes “Much of Exadata’s upfront cost is for software licenses, and those are perpetual licenses you can reuse even after your hardware is obsolete.” But there are two flaws in that argument:
- Chargeable upgrades — Oracle likes to charge for a new major release every few years
- Moore’s Law improvements being instantiated via core count — CPU clock rates have stabilized, and so almost all processing power improvements now come through core count. Oracle commonly charges on a per-core basis.
Obviously, if you have an “all you can eat” long-term enterprise Oracle license, covering licenses and maintenance alike, that’s a different matter. But I don’t think there are very many of those.
Comments
19 Responses to “Sorting out Netezza and Oracle Exadata data warehouse appliance pricing”
Leave a Reply
[…] Netezza claims this will all produce an immediate 10-15X increase in price-performance, based on a 3X cut in price/terabyte and a 3-5X improvement in mixed workload performance. (Edit: Actually, that ’s not true using the ordinary meaning of “price/performance”.) […]
@CURT: “That’s a little under $100K/terabyte uncompressed, vs. Netezza’s figure of a little under $45K uncompressed. I would guess Oracle’s compression is a little better than Netezza’s, but only a little.”
When looking at price/performance, I’m guessing the the Oracle compression has a pretty decent impact to performance (CPU cycles wasted on compression not doing SQl work). TwinFin, using FPGAs, should have zero impact to performance with compression running. Performance might actually go up? Do you agree?
Re:Pricing
Maybe with a handful of the “new guys” publicly stating our pricing. We can help you force the trend of explaining how companies market their products.
I know our launch got NZ’s attention. If you read Phil’s BLOG (http://www.netezzacommunity.com/blogs/nzblog/2009/07/30/catch-a-wave-and-youre-sittin-on-top-of-the-world). He clearly mentions Dataupia and XtremeData as the competitors that helped set the new $20K / User Data TB mark.
I’d like to note that we are hitting that mark of $20K/TB User Data – without compression. (Vs Oracle = $100K and TwinFin = $45K).
dbX with compression… coming soon
My guess is that your pricing model on Netezza is likely backwards. Based on the pricing of the old 10000 series, compression was an additional cost. With the old systems you’d pay (GSA Schedule 70 price) around $1mm per rack and an additional $500k per rack for the compress engine.
So my guess is that you probably pay ‘around $20k’ per TB of uncompressed data. Considering that the disks are 2.5X bigger than in the 10000 series that doesn’t seem out of line to me. If you want compression then you would have to pay an extra cost for the compression feature… based on the 10000 series that would be around 1/2 the cost of the original system without compression but I suspect might be price higher in the TwinFin models.
So I’d guess that the TwinFin 12 (32TB of user data) is priced around $20k * 32TB = $644,000 without compression. Compression probably costs around 1/2 of that price, bringing the total to around $1mm per rack. This is just a complete guess though, I’ve had no contact with Netezza about pricing.
It is also possible that Netezza has decided to get rid of the extra charge for compression and just include it in the base package. If that is the case then their $20k per TB pricing could be assuming some standard compression rate like 2.25x or whatever. It is possible they went this route but it would be a departure from their pricing model with the 10000 series.
I believe compression is standard on the new Twin Fin boxes .
-Dave
I understand why you object to the “10-15x price/performance improvement” claim, but you’ve gotta give the Netezza marketing folks a break.
To give a marketing department raw material like “3-5X performance improvement and a 3X decrease in price/terabyte” and then tell them NOT to multiply those numbers together (not to mention then round 9 up to an even 10)… that’s asking too much! Might as well ask the Sun not to rise in the East.
@Curt
Do you know of any data points on Netezza compression other than what the marketing group gives out as a general range (2-4x)? I’m especially curious as compression wasn’t introduced until release 4.5 (May 2008) and it is an additional pay for option which is off by default so I would think it is not widely used. Given the limited public information on Netezza’s compression, I think any comparison would be a guess.
The Oracle figure is “user data capacity” (as quoted from the data sheet), so what ever objects (tables, indexes, etc.) are created occupy space in that storage pool, if you will. However, there is really is no need for Oracle to specially account for space for indexes as they are not required for performance with Exadata. In fact, most of the POCs that I have been involved with require zero or very few indexes (which are used simply for single key lookups).
@Geno
I’m guessing you haven’t done your homework on Oracle compression 😉
From Table Compression in Oracle Database 10g Release 2:
@Greg Rahn:
The POCs you have been involved with Exadata have used any kind of compression?
If yes, which one? The standard, available from 9i or the new 11g Advanced Compression?
@Bence Arató
They’ve used COMPRESS FOR DIRECT_LOAD OPERATIONS which is the one available prior to 11g Advanced Compression (COMPRESS FOR ALL OPERATIONS) because the loads have all been bulk direct path operations.
Greg,
How about the production Exadata sites you’ve been involved with or otherwise know of? Are they also all but index-free?
If you don’t actually know of any production Exadata sites, I withdraw the question. 😉
Thanks,
CAM
@Curt
The production Exadata Database Machine sites that I am aware of are very index-light to use a term you are familiar with.
This should really be no surprise if you understand the physics behind single block/page table access via an index versus parallel large multi block/page table access via full table scans.
Curt:
I’m interested in your claim that ‘Oracle likes to charge for a new major release every few years’. My understanding is that Oracle licenses migrate to new releases without charge and that this is actually a key benefit compared to the likes of Netezza who force customers to repurchase both hardware AND software.
Have I misunderstood the different licensing models?
If so, under what scenarios would Oracle charge customers to upgrade to a new major release? (besides, of course, purchasing additional options)
Craig,
http://blogs.zdnet.com/BTL/?p=5942 probably puts it better than I did.
When does Oracle add additional charges for new database features? Any time they need more money to pay Larry Ellison’s $100+ million salary…
Curt:
After dealing with Oracle compression for a few month I gave up. Even though 11g compression performs slightly better, under the scenarios we tested and run, even though Oracle compresses 4x, the performance is at least 25% worse than uncompressed data. Based on that I came to the conclusion that, on Oracle, compression is only to save space on historical/stale data.
Netezza spends 50% of its time on disk read as it is brute force read type machine, and I am pretty sure that 3x compression means about 50% increase in performance for them with FPGA in the background. Also the rumor is that they are introducing intel chips in the SPUs along with the FPGAs to do more complex stuffs more efficiently. I don’t know about the price performance, but from their sales guys I heard first hand that just performance would improve anywhere from 5-15 times.
If I have 100K to spend per TB, I will always ask myself why should I go for Oracle/Exadata, when Teradata is there!
Aniruddha
Curt,
The zdnet.com article definitely puts it better than you as it isn’t misleading. It makes it clear that any charges are for the use of new features and not just for upgrading. Your article suggests that Oracle charges for upgrades alone which simply isn’t true.
I would imagine that one of the reasons why Oracle claims a more benign long-term pricing model is because unlike Netezza (or Teradata) they do not force customers to repurchase software they have already bought during upgrades.
‘Oracle likes to charge for a new major release every few years’.
This is 100% false. As long as you pay support, you get upgrades, for free, forever.
@Terry & Craig
Sorry Guys, I agree 100% with Curt on the paying for Oracle Upgrades.
Paying for Oracle support does not give you free upgrades. You are paying for it. And if you look at the cost breakdown for an oracle licence renewal, you will see that is consists of 2 items: Technical Support and Upgrade Fee.
So you are indeed very much paying for an upgrade to a new release. And this is before you start adding in the cost of new options.
Of course, if you don’t pay an annual licence renewal, then if you want to upgrade to a later version of Oracle, then you will need to pay for a whole new software licence. Or if you are in good with Oracle, they may let you pay back maintenance on your existing licence.
Also note, that you have excluded the cost of traditional hardware replacement in your upgrade costs. We usually upgrade our database server hardware at around the same time as we upgrade the software version. So even though Netezza may require a full hardware upgrade, wouldn’t you be doing that with a traditional platform, or exadata anyway?
@Curt
My company is currently reviewing options of improving performance of our existing traditional EDW (which happens to run Oracle on PA_RISC HP-UX). And we are considering an appliance as part of our hardware LCM which requires a hardware replacement. Your price/performance comparison is very helpful. Thanks.
Neil,
You’re most welcome!
I was sloppy in that, if one pays maintenance, there indeed is always another version one can upgrade to w/o additional license fee. But there’s no assurance that all new functionality will be in that version; it could also come in the form of chargeable options.
E.g., a free upgrade to 11g would hardly inoculate one against paying new license fees for all the server-tier stuff in Exadata.
[…] Oracel Corporation will someday consider whether to continue charging customers for a major release every few years, as discovered by industry analyst Curt Monash. […]