blisenko -
I see you already have several detractors on your post. But META is a pretty good group... and their model shows what most of us have seen in the ERP space - and also back end services for web farms. Both of these system designs can scale almost infinitely on the front end, and the tendency is to have the lowest-cost systems possible in that slot. But overall scalability is determined by the back end - and in the case of SAP, the data does not partition well, so a big honker database machine is what determines overall system headroom.
META's numbers include the startup costs for the SAP stuff - which are typically many times the rest of the capital costs.
In an SAP system, there is some benefit to having a common administrative model, so given that the customer wants headroom and selects say a UE10000 as his database machine, he is likely to pick a Solaris capable platform for his app servers as well - and is probably money ahead to do that, even if the base hardware cost is twice the level of an NT based solution.
Intel based systems have gradually been moving up the food chain in ERP, but mostly at the expense of HP - Sun still has a BIG headroom advantage that assures them a place in the larger ERP implementations.
In web server back end design, the data is often a lot more partitionable - and there, the version of SQL Server that MSFT and CPQ used to run the 200K TPC benchmark would work quite well to provide headroom. That may give them a shot at some of those big web site back end opportunities later this year when the product ships, especially if the database has reliability built in, as most partitionable systems do. The numbers are compelling - nearly unlimited scalability, and at 1/5 the price of a UE10000 / Oracle system which tops out at around 100K transactions.
But in the ERP space it is the big single system which wins, and UE10000 is still king there. I don't see anything on the horizon to change that for at least another couple of years. |
