Chapter 3. Hardware Architecture

With the software architecture properly defined, the next biggest challenge for the data warehouse DBA is to select an appropriate hardware platform for implementation.

In theory, data warehouse hardware selection should be simple. Data warehouses are huge, so common sense would dictate ordering large, scalable systems. Once you start throwing around the "T" word (for terabytes) when referring to your data warehouse, then lots of CPUs and lots of disks should not be a hard sell. If you need further convincing, then reexamine your data warehouse's mission statement and sponsor. A data warehouse enabling both executives and senior management to form strategic business decisions is worth a lot to any business. Hence, you would fully expect a hardware budget in line with that supposition. In short, data warehouse platforms are not cheap.

As simple as I've tried to make the hardware budget and selection process sound, nothing in life works as simply as we think it should. It's not unusual for many mid- to large-sized companies to have preferred hardware and software vendors who typically have early and direct access to project requirements. Furthermore, such vendors often provide streamlined ordering, price breaks, and other perks to attract and maintain key accounts. As such, it often occurs that hardware decisions have already been made by the time the DBA becomes involved with the data warehouse. Don't let this happen, because even the best software architecture will fail on the wrong hardware.

Another challenge with hardware selection is the rapid pace of technological advancement. Even the people selling the hardware have a tough time keeping up with just their company's offerings. The poor DBA often must serve as an expert across the various hardware vendors and their offerings, such as servers, disk arrays, tape management systems, etc. Add to that all the related software and it's no wonder that this phase can leave many DBAs stressed and second-guessing. Of course, between the vendors, business sponsor, technical management, developers, DBAs, and system administrators, there will be no shortage of well-intended advice.

The final challenge is to pick a platform that has a committed growth path from the vendor. With IA-64 and other new technologies, some RISC architectures may not have a simple and straightforward growth path. There may be cabinet, board, or bus swap-outs, and possibly even OS switches required. Press the vendors extremely hard on this issue because whatever hardware you buy must be scalable both in terms of database size and concurrent users.

If as the DBA you are lucky enough to be involved from the start, review all the technological offerings and find a hardware platform with a well-defined growth path?it's safe to bet that it will be the hardware vendor's premier equipment. No one said that a data warehouse would be cheap! If the budget is tight, do not compromise on lesser equipment. The cost to replace the wrong hardware selection is generally more than that of the delay to secure additional initial budget. Remember, a data warehouse can be delayed. It is not a core business system like OLTP and ERP applications (i.e., a business can go on without a data warehouse), so wait for sufficient funding.