Event
Summary
July
6, 2000 [Source: IBM Press Release]
IBM,
Intel and Microsoft announced the world's fastest server cluster for commercial
use, recording performance levels that triple the performance of Oracle
running on a Sun Microsystems cluster, at one-third the cost per transaction.
Using
the performance measurement technique agreed to by all computer makers
(TPC-C), IBM Netfinity and DB2 Universal Database, Intel Pentium III
Xeon™ Processors and Microsoft's Windows 2000 achieved 440,879 Transactions
Per Minute - record-breaking results in server and price performance.
IBM,
Intel and Microsoft joined forces on this groundbreaking effort to prove
that a combination of Netfinity Servers with Pentium III Xeon (TM) processors
running at 700 MHz (megahertz) with 2 MB (megabyte) L2 cache, IBM DB2
Universal Database and Microsoft Windows 2000 Advanced Server operating
system provides a highly scalable environment. This technology combination
is ideally suited for data-intensive applications like business-to-business
(B2B), e-commerce, and Enterprise Resource Planning (ERP).
Market
Impact
These results affect the server market in a few ways:
From
a hardware standpoint, IBM's Netfinity can claim the TPC-C crown by
a stunning margin. Comparing it to the best results from Sun Microsystems,
the Netfinity handled 3.25 times the number of transactions per minute
of Sun's cluster, for only slightly higher cost (if $14M vs. $13M can
be considered slightly higher). This translates to the Netfinity having
a $-per-transaction cost one-third that of Sun - another big win in that
area. When comparing these results to the previous best - Compaq's recently-withdrawn
ProLiant 8500 results for 64- and 96-CPU clusters - there's an apparent
near-doubling of performance, but the price/performance appears worse
($32.28/tpmC for Netfinity vs. $19.12/tpmC). However, these results were
withdrawn because Microsoft's SQL Server 2000, which Compaq used, violated
some of the testing rules [1]. Thus, it is unclear
whether a re-test of a Compaq cluster will provide equivalent results.
From
a software standpoint, this is potentially a big win for IBM and DB2,
and potentially a setback for Microsoft and Oracle. The positive for DB2
is obvious: bragging rights for the highest performance. In addition,
these results push DB2 into the limelight, a new and welcome experience
for IBM in this area. The potential downside for customers is that DB2
is not cheap, costing more than $20,000 per CPU, not per server. (Note:
Oracle is also expensive, SQL Server's price is significantly lower than
both Oracle and DB2.)
The
negative for Oracle is that they no longer own the high-end database performance
marks. They lost that distinction in February when Compaq used SQL Server
2000, but the recent withdrawal of those results put them back on top
- for three days.
The
negative for Microsoft is the embarrassment of Compaq's results getting
pulled because of the SQL Server problem. Microsoft can reverse this embarrassment
by posting similar results with the bug-fixed version. (Microsoft has
reportedly fixed the bug/issue, but the fixed version has not yet been
released.)
[1]
Clause 1.6.3 of the TPC-C specification, which requires that "any set
of rows or columns" in the TPC-C database be able to be manipulated using
"general purpose mechanisms."
User
Recommendations
IBM should now join Oracle on a customer's short list of large-transaction-volume
database software. DB2 has always been a high-end, quality database (as
is Oracle), but it has rarely received the publicity and mindshare that
Oracle has commanded. (Perhaps if CEO Louis Gerstner flew his own Gulfstream
into San Jose's airport after curfew?) Although one victory does not necessarily
lead to major market share, this demonstrates that DB2 is moving beyond
its traditional bases of IBM's S/390 and RS/6000, into the Windows-based
world. Because DB2 is "seamless" across the enterprise, this means future
expansion is not precluded. The relatively low per-transaction cost means
that customers not committed to Sun/Solaris should seriously consider
DB2.
The
downside, as mentioned before, is DB2's relatively high price. The post-34%-IBM-discount
pricing for the tested system still had the DB2 licenses running about
$2 Million, versus less than $1 Million for SQL Server [Source: TEC estimate].
If IBM decides to drop the price even further, this product would give
both Oracle and SQL Server 2000 a big challenge. Even without an additional
price cut, the price/performance figure makes it a good competitor to
an Oracle/Sun setup.
If
customers have a large operation and the cash to support it, IBM's DB2
UDB is definitely a strong choice.