datawarehouse appliances

Hi,

I would like to know in which cases are recommended tu use a datawarehouse appliance solution like IBM Netezza or Oracle Exadata.
I know that this kind of solutions are very expensive and used in a very large datawarehoses, but I would like to know
if there are more criteria in order to assess if you need a solution like that or yo can go with commun databases, servers, etc..

Thanks,

Juan

I have worked on Netezza systems for the last few years, so I can speak of them. I have not used an Exadata system yet so I will not comment on that.

Netezza systems are delivered and installed pretty much like an appliance (the source of the term). It is a fully contained and configured system that is delivered, plugged in, and ready to use. You access it and develop databases like any SQL database (via OBDC, JDBC or CLI).

It is a massively parallel, shared-nothing architecture. A system can have as few as 4 to as many as 1920 cpus each with their own dedicated disk. All disk is mirrored and there are spare CPUs in the system to handle hardware failures without interrupting running queries.

It is a very fast system and was designed specifically with data warehousing in mind. It is a brute force approach to computing. The databases are very easy to manage as the system does not support indexing. All queries are essentially table scans. The system maintains an internal structure call zone maps that it uses to skip over disk pages it does not need (based on predicates in the query) and most of the processing is handled by a field-programmable gate array (FPGA) processor that is in direct control of the disk. The CPU only sees the data (columns,rows) it needs to see and not burdened by the drudgery of loading, parsing and evaluating disk blocks.

Development and support are much easier than on a traditional system. Without indexes, there are only two decisions (distribution and organization) that you need to make when implementing a table, and then tend to be easy choices. The DBA's life is a lot simpler as well, needing to an occasional GROOM and statistics to keep things running well. Both of these can be done while the system is live.

So, the notion of cost needs consider the lifetime of the system. A Netezza system is not expensive for what you get. Being a DBA on medium sized system is not a full-time job. In very large shops with dozens of systems, the DBA staff is a fraction of what an equivalent Oracle staff would be.

Thanls for your help, ngalemmo

Exadata is just Oracle with smart disk management, i.e. only pull columns from disk that are in the query.

On-premise appliances are a dying breed. They're not dead, yet, but they're in palliative care.

The reason is that (a) on-premise SMP with SSD storage is catching up on performance, and (b) cloud MPP offers better performance at a fraction of the price.

I used to lead sales of MPP on-premise appliances, across 1/3 of the world, for a major vendor. Our typical sale, including HW and SW, was between $500k-750k.

These days you can get equivalent performance and scale on Redshift, Snowflake or Azure SQL Data Warehouse for less than the annual maintenance and support of that on-premise solution.