Salvatore Salamone
When running large database applications, mainframe computers have traditionally offered impressive performance advantages over their client/server counterparts. Not any longer, say Digital Equipment, Oracle, and analysts. Such high-end network servers as Digital's new AlphaServer, when combined with new 64-bit RDBMS (relational database management system) software, now offer an alternative to mainframe systems. But the success of 64-bit database servers will rely on a number of factors: Don't expect to see a mass migration from the mainframe any time too soon.
A 64-bit system can deliver a performance improvement of between one and two orders of magnitude over a 32-bit system, according to analysts at the Aberdeen Group (Boston, MA, (617) 723-7890), wh
ich makes client/server database performance comparable to that of many mainframe systems. The combined efforts of Digital Equipment (Maynard, MA, (508) 493-5111) and Oracle (Redwood Shores, CA, (415) 506-7000) mark the first entry into this 64-bit area, which Aberdeen calls the LIMD (Large-scale In-Memory Database ).
Oracle rewrote its 32-bit RDBMS, Oracle7, into a 64-bit version called Oracle VLM (very large memory). Oracle VLM, when run atop an OS like the 64-bit version of Digital Unix, lets you load much more of a large database into memory than is possible with a 32-bit program.
Typically, 32-bit systems are limited to main memory database usage of 2 GB. With Oracle VLM, the amount of addressable main memory is essentially unlimited, according to Larry Ellison, Oracle president and CEO. For example, Digital's new AlphaServer 8400, an enterprise server, and the 8200, a departmental server (both of which are the first to use the 64-bit, 300-MHz Alpha 21164 chip), can support up to 14 GB of m
ain memory.
Additionally, compared to the 32-bit version, Oracle VLM takes fewer operations to move the same amount of data into memory. Oracle7 allows data transfers from disk to memory in 2 KB blocks (8 KB-size transfers were supported, but Ellison says that size was seldom used). In contrast, Oracle VLM moves data from disk into memory in 32 KB blocks (called Big Oracle Blocks or BOB for short). The increase in performance is compelling (
see the chart
).
The acceptance of 64-bit database systems, whether from Oracle or some other database vendor, depends on several factors. One important one is the availability of less expensive memory, which should start occurring in late 1995 with the arrival of low-cost 16-Mb memory chips.
A second factor is the mainframe's lower support cost per user. An official at IBM points out that when companies like Oracle tout the price/performance advantage of a client/server solution to a mainframe, they are only quoting cost of purch
ase and neglect to mention the hidden costs of user support.
A 1994 report produced by Forrester Research (Cambridge, MA) says that it costs three times more to support a 5000-user network ($6.4 million a year) than it does to support an equal number of users on a traditional IBM mainframe ($2.3 million). A third factor is the pace at which all applications that an organization typically runs on a mainframe--including administrative, financial, and other mission-critical applications--are ported to a given 64-bit application.
Aberdeen's analysts say it's not likely that mainframe customers will abandon their big iron unless their important applications--not just their databases--can move to client/server systems.
illustration_link (10 Kbytes)
The benchmark tests were performed on a Digital AlphaServer 8400 and used five database tables ranging from 70 KB to 5.2 GB in size. The performance gains of the 64-bit Oracle are due to the system processor's ability to directly address much more main memory than 32-bit systems, as well as being able to move data into memory in larger chunks. Tests measure time in seconds to perform transaction. Smaller numbers are better.
Oracle Releases 64-Bit Database Bullet
