It has been said that the job of the analyst is often to comfort the afflicted and afflict the comfortable. IBM’s recent series of announcements related to a new generation of mainframe should leave us comfortable about the mainframe’s future and IBM comfortable that it has put its best foot forward by providing a breathtakingly comprehensive and effective upgrade of the mainframe’s capabilities. Therefore, it’s time for a good analyst to step up and afflict the comfortable with niggling caveats about areas where IBM could have done even more.
Specifically, IBM could and should have done more in two areas: columnar technology and Windows workloads on the mainframe.
Let’s start with columnar. For the typical enterprise, Business Intelligence- (BI-) type data management costs too much and frequently performs less effectively than it needs to. Data in existing data warehouses grows by 50 percent per year due to compliance and regulatory requirements and enterprise demands for storage of new types of data. A business environment marked by a high rate of change calls for more frequent analyses performed on fresher, time-critical data. As a result, data storage and workloads are growing faster than hardware performance and price-performance improvements.
In many cases, columnar technology—storing relational data as columns instead of rows—has proved to be an excellent answer to these problems in the largest of Very Large Database (VLDB) implementations (see, for example, recent Sybase IQ and ParAccel 1- and 3TB TPC-H results). The reason is that columnar databases can compress the data and indices, sometimes by an order of magnitude, compared to relational, and carry out transactions on the compressed data. The result is often order-of-magnitude improvements in performance vs. a comparable relational database. Columnar technology is now well-proved: Sybase IQ has been at it for a decade and it’s widely available; smaller vendors such as Vertica offer it, and Oracle has added a columnar option to its database appliance, Exadata.
However, even in the new zEnterprise, IBM lacks a distinct, full-fledged columnar database. Even if the user acquires Smart Analytics Optimizer, IBM apparently applies columnar technology to only those parts of a data store that it deems appropriate, with little user control over fine-tuning the columnar operations. The result is that a major source of BI performance improvements will not be fully exploited, nor be an equal partner in the mainframe’s highest-scaling BI solutions.
Now about those Windows workloads. As I noted in a post a year ago, since the advent of Linux on the mainframe, there has been a glaring gap in IBM’s ability to support workloads on the mainframe—the lack of support for Windows.
There are clear and cogent reasons to want to move Windows workloads to the mainframe, including the mainframe’s superior energy profile. Another reason is the ability of the mainframe to deliver long-term costs per app well below Wintel scale-out in cases where the enterprise can move 20 or more Windows apps to a single mainframe. More urgently, the aim of today’s cloud computing implementations is to cut large-scale costs both of administration and of per-app hardware; and the mainframe excels at both. But as long as the mainframe can’t support a large body of enterprise Windows apps, external cloud vendors or internal cloud creators will have to decide between a partial solution that replicates the existing Wintel/mainframe split and doesn’t save much in the way of costs, and movement toward Wintel that will increase long-term costs.
It’s now possible to talk about the mainframe handling most apps inside or outside a cloud and to envision migrating most Windows apps to the mainframe in the lifetime of an IT administrator. Mantissa Corp.’s z86VM offers emulation of Windows over z/OS. Novell SLES Mono Extension allows users to port a large proportion of today’s Windows apps (by some estimates, 50 percent or more) that depend entirely on .NET without the need for code change or conversion to a different language. Meanwhile, as an IBM speaker recently noted, an increasing proportion of mainframe users are telling IBM they would like to put Windows workloads on the mainframe.
Now that IBM has capped a five-year string of major mainframe upgrades with a new z generation, it’s time to hold the mainframe to a higher standard. Doing something more with regard to columnar technology and Windows workloads on the mainframe would be a clear indication that IBM can meet that standard and deliver major benefits to mainframe users.