Mikey - The benchmark was run on a cluster of 8-way ProLiants. CPQ developed the technology for the 8-way system with Corollary. Before the work was complete, Corollary was bought by Intel. CPQ retained some of the rights for their work - which means that anyone who sells an 8-way box pays CPQ a royalty, directly or indirectly. But more importantly, Intel's version of the 8-way product was plagued with problems. DELL and others who designed boxes using Intel's version were unable to actually sell any, leaving the market for volume 8-way servers exclusively to CPQ.
CPQ has taken good advantage of their 9-month window, placing 8-way machines into large accounts, and with ISVs like SAP, to get people to develop for the big machines. One of the most spectacular fruits of that work was the effort with MSFT around the next generation database and OS products which resulted in the 227,000 TPM-C benchmark.
The notion that CPQ somehow "cheated" on this benchmark is just sour grapes - if that were true, then IBM, HP or SUNW (who are all represented on the TPC council) could have protested the result. Every detail of the test is published and any member can challenge the result if there is a suggestion of hanky-panky. Every MSFT result gets close scrutiny since the TPC council is predominantly old-line big-iron people.
CPQ now has 4 of the top 10 performance numbers - 2 with MSFT SQL Server, and 2 with Oracle 8, 3 with ProLiant and 1 with Alpha. I would imagine that there will be some new big numbers with Wildfire, also on Oracle (since SQL Server does not run on Alpha). see tpc.org SUNW has 2, IBM has 2 (and 1 with BULL), HP has 1. DELL does not have an entry in the top 10.
In the price-performance race, the "big iron" guys are out of the game - MSFT SQL Server has all of the top 10. See tpc.org Here DELL has 3 entries, CPQ has 4, HP has 2... DELL has the only 8-way entry in the price-performance list. Although they published their number in November of 1999, they still have not resolved all of the issues to actually ship their 8-way products, and CPQ's single-system 8-way number is higher, although at a higher cost.
TPC allows results to be published using components which will ship within 6 months, and DELL is taking advantage of that with their number (as is CPQ on the big cluster benchmark, since the new MSFT SQL product will not ship until May or June).
DELL theoretically could put together a cluster benchmark but they have never done one - perhaps because they don't sell database clusters.
TPC numbers do not include shipping and some council members have suggested that DELL's price-performance numbers are artificially low because they put some of their profit into "shipping". Since there is no other way to get a DELL server, this allows DELL to get a few percent advantage in cost for the test, although "landed cost" (what a customer actually has to pay) might be higher for DELL. But under the rules as they exist today, DELL does not have to include charges for shipping.
This is an old trick in the direct marketing business. When the "Danbury Mint" and the "Franklin Mint" were competing hard on price for special coins (these are not real coins but are produced just for the "Mints" and are not legal tender), that got taken to the extreme. Franklin offered a coin for $99.99 plus shipping and handling - which came to $4.95, so the total cost to the customer was $104.94. It actually cost them $.50 to ship the coin.
Danbury countered with their offer - only $79.96, divided into four easy payments of $19.99 plus shipping and handling (yep, $4.95 per payment, adding a total of $19.80). Cost to the customer was $99.76, about $5 cheaper than the Franklin coin, but the "offer" sure looked like it was $20 cheaper. And the "real" cost of shipping was $.50 in each case.
But the "shipping and handling" costs on DELL servers are also an area that savvy corporate buyers are working. One guy told me he closed a deal with "OK, you've got the business, but FOB your dock - we'll pick it up." This makes invoice price and landed cost the same... |