How does Yellowbrick compare to Netezza in our Benchmark?
Yellowbrick makes some very bold statements about how much faster and cheaper it is than other data warehouse platforms.
When it comes to Netezza, Yellowbrick has a strong claim to being its natural successor. Customers migrating from the Netezza platform will find it very simple. This is because from the get-go Yellowbrick made sure that they had a high level of compatibility with Netezza SQL functions and even their load utilities mirror those of Netezza.
Yellowbrick says it is up to 100 times more performant than Netezza. We decided to put it to the test and benchmark Yellowbrick against our own Netezza system, and the results make good reading.
In the benchmark we modeled the sales and return process for a retail business with store, catalog, and web sales channels. We used a 36GB database* and 17 dimension tables. The largest fact table had 400 million rows. There was no physical database or workload management optimization/tuning.
We simulated a mixed workload (465 unique queries) across 11 concurrent sessions with reporting, ad-hoc, interactive OLAP, and data mining type queries of varying complexity.
* TPC Benchmark™ DS - Standard Specification, Version 3.2.0, June 2021
Copying is by permission of the Transaction Processing Performance Council
© 2021 Transaction Processing Performance Council
What we compared
Netezza: N2001 PureData for Analytics Netezza Striper instance containing: • 4 nodes • 128 vCPU • 512GB of RAM
Yellowbrick: small-v1 Yellowbrick Cloud instance on AWS (m5dn.4Xlarge) containing: • 4 nodes • 64 vCPU • 252 GB of RAM
It’s hard to compare Yellowbrick pricing to that of Netezza because unlike Yellowbrick, IBM is not transparent with theirs. So, we used the actual historical list pricing and also a heavily discounted price of 70% which some customers may be able to negotiate for volume or long-term commitment. If you want to do a comparison with your own system, share your pricing with us, and we’ll do it for you.
Netezza Hourly Cost
|List Price Assumption||70% Discount Assumption|
|Maintenance/support for years 2-3||$250K p.a.||$75K p.a.|
|DC hosting (power, network, etc.)||$30K p.a||$30K p.a.|
|Hourly cost (Total Cost/3/(365*24)||$70 p.h.||$23.40 p.h.|
Yellowbrick Hourly Cost
AWS Charges: Shared Services nodes that host the Cloud Data Warehouse Manager (CDWM) control console, DWH Instance, and other supporting services e.g. for system monitoring, bulk loading, etc. There’s no software license charge for these, but the fixed AWS costs of running them all full time amounts to around $1.86 p.h. (although some nodes can be suspended when not needed to reduce this).
3-year m5dn.4Xlarge reserved instance virtual compute cluster: $0.423 p.h. X 4 nodes = $1.69 p.h.
S3 Intelligent - Frequent Access Tier: $23/TB/month X 23.4TB = $538.20 p.c.m./730=$0.74 p.h.
Yellowbrick 3-year Subscription License: $0.055c per VCPU, p.h. X 64 VCPU = $3.52 p.h.
Total hourly cost = $7.81
- Yellowbrick finished in just over 3 minutes.
- Netezza took over 3 hours to run the same queries**, with one of them failing completely.
- Netezza was 10X slower per query.
- Yellowbrick 50X better throughput.
- Netezza 167X more expensive.
** The longest running Netezza query:
- Reported the top reasons and average quantity,
value, and cost of sales returns for different customer and sales types via the web channel.
- Medium Complexity
- 8 way join (including one self-join)
- Aggregation, Grouping, Ordering
- Took 1 hour 20 minutes to run on Netezza; 9 seconds on Yellowbrick
- Yellowbrick is quantifiably orders-of-magnitude better in terms of price, performance, throughput, and scalability than the alternative.
- If you’re…
- still on a TwinFin, Striper, or Mako generation Netezza appliance,
- thinking of migrating to something else, or have already done so but are unhappy with your choice, and
- interested in learning more and seeing what Yellowbrick and Smart Associates can do for you,
- …then please reach out and get in touch.
To see our CEO Huw Ringer running the benchmark click here.