THE DATA WAREHOUSE ON S/390: PART 1
By Dan Graham, global strategy and operations executive,
Global Business Intelligence Solutions, IBM; and
Malcolm Nolan, marketing manager, Business
Intelligence, S/390, IBM
Why run your data warehouse on a mainframe?
In this issue we are going to discuss the role of mainframes as a resource for data warehousing. On that note we can already hear some of you thinking: Uh oh, these IBM guys are harking back to the glory days when mainframes ruled the roost.
Don't rush to judgment. We are not talking history. We are as up to date as today, and as forward looking as tomorrow. Here's why.
In July 1988, the Palo Alto Management Group surveyed 374 customers who operated data warehouses. PAMG asked them whether some or all of their operational data was stored on a mainframe - 75 percent said yes. You can massage the figures in a number of ways but we think it is fair to say that 75 percent of warehouse operational data resides on mainframes.
So the first point to make is that three quarters of warehouse-relevant data already resides on mainframes. The second point is that mainframe users have already invested in infrastructure, application software and support systems, so everything needed for business intelligence (BI) is in place on the machine where the data resides. And finally, there is no reason to ship data around at networking speeds when it can be done at Escon speeds in the place where it lives.
So why would CIOs run data warehouses on something other than their mainframe operations server? In previous years, nobody had the technology to cope with operational systems and data warehouse requirements in the same breath. Under those circumstances the best fire wall protecting operating systems was to segregate the insatiable appetites of the warehouse and its user community. But times have changed. We will come to this later.
What it amounts to is that a reliable infrastructure, existing mainframe skills, and characteristics such as performance and security - all integral elements in the S/390 world - can be counted on to run BI workloads. This yields a higher return on installed investment (ROI). In effect, adding the warehouse function to the S/390 is the equivalent of getting higher mileage from your car.
Meanwhile, even as the tried and true mainframe is leveraging ROI, in a real sense it is also serving as a reliable low-maintenance system for the CIO who has enough complexity already.
The best way to reduce complexity is to eliminate risk: Stick with the medium you know. We have both seen situations where a shop has installed something exotic and nobody, but nobody, has the skills to make it purr.
The problem with the latest fashion in servers is that once they are up and operational, the next challenge is system outages and downtime when BI was a novelty, users and managers were fairly tolerant of downtime. But, as BI becomes mission critical, and as operational systems come to rely on it - to the point where BI may drive operations - so the margin for downtime is shrinking fast. Most servers running BI loads can experience downtime periods of up to 438 hours per year.
Contrast that with S/390 processors, where the average downtime is five minutes a year.
When you put warehousing on any platform it becomes the immediate focus for a multitude of users requiring a wide variety of different applications. The beauty of S/390 is that, years before data warehousing entered the IT stream of consciousness, mainframes were already performing a wide array of user demands. The system responded by evolving a huge network of facilities and operational characteristics. To keep the many elements in step, IBM created Workload Manager software which, historically, had nothing to do with data warehousing. It had everything to do with Customer Information Control System (CICS), IBM's transaction processing monitor that runs the majority of Online Transaction Processing (OLTP) in the world today. Workload Manager regulates the balance among batch workloads, interactive users, transaction and Web serving applications.
Its main task, historically, was to ride herd on those subsystems and make sure nothing stepped out of line.
Tests at IBM's S/390 Teraplex Center showed that customers could - using Workload Manager - allocate specific users to query classes and guarantee response time for some of those classes. Meanwhile other query classes run quietly in the background.
The key point here is that the DB2 and OS/390 communities have worked together for a long time to solve complex issues. Initially those issues had nothing to do with data warehousing, but we have found that this marriage of minds makes the combination a winning solution for emerging challenges confronting CIOs running data warehouses. We can safely say that the decades of research IBM invested in DB2 and OS/390 lets BI run almost transparently on an S/390 Parallel Enterprise Server.
[In our next installment: Growing the S/390 system; partition perfection; the total cost of ownership; and integrating the warehouse with operations on a single server for leveraged ROI.]