What are statistical performance profiles for RDB technology on NAS?
Companies like NTAP are saying that running a relational database on NAS yields acceptable performance and operation. But, they never say that to DBAs where we need to know about what are the transaction frequency and intensity boundaries, schema issues such as locking where non-linear relationships are used (i.e. foreign table pointers, multi-dimensional databases and spatial objects/databases), row and column size scale issues, mount times for different size databases (especially large, multi-terabyte) and numbers of mounts per mount point.
There's a very standard answer when people ask performance questions, "It depends." That seems trite but it illustrates that there are an incredible number of variables and environments that can have impact on performance.
You've brought up a few specific characteristics that are important to consider for the type of operation that you do. There are also many others including network congestion, size of server, etc. Probably the most important one when considering OLTP is the expectation of the customer. What is the required response time? Many believe that in enterprise-class operations, response time has to be below a certain level (guaranteed) to be acceptable and NAS boxes may not provide that so they typically use direct attached or SAN attached storage. There are two things we generally recommend. Contact your database vendor and get their recommendations. They have performed many environment tests and can give you guidelines for what to expect given certain equipment, database size, etc. You need to get to their group that has the performance information (beyond the salesman). Oracle does a very good job at this.
The other recommendation is to get a performance guarantee (in writing) from the hardware vendor. If they say it will meet your needs, hold them to it with a guarantee that if it doesn't, they will do whatever it takes, including installing other vendor's equipment, to make it right. If they are going to be aggressive with their statements, they should back it up. Unfortunately, many salesmen don't understand their customer requirements and environments very well and may not give an accurate recommendation.
This was first published in April 2001