Re: Sybase MPP shows excellent scalability to 128 CPUs

NRaden wrote: > I was not over generalizing at all, I was referring to all of the TPC > benchmarks. The TPC-D benchmark is highly suspect in my mind. What > I can draw on is 15 years of experience developing complex decision > support systems and 4 years directly involved with data warehousing. > > Speaking from experience, TPC's have been completely useless as an > indicator of performance in this environment. The apples-to-apples > analogy is falacious. DWs do not insert online, indexes are not > updated, queries are large, caching characteristics are completely > different. I agree completely: currently available TPC results are an abysmal indicator of DW capabilities. Of course, currently available TPC results come from OLTP workloads. TPC-D represent a marked departure from existing TPC workloads. Queries *are* complex. Cache utilization patterns *are* different. Engine stress points have moved from the access methods of OLTP, with singleton-row retrieval and concurrency concerns, to SQL optimizers and expressions evaluation and promotion. No one would claim that TPC-D represents the final word in evaluting a system's ability to handle decision support. Very few people, I hope, would dismiss it out of hand becuase it springs from an organization whose focus, to date, has been on OLTP workloads. And I, for one, think it can be a valuable start in providing public, defensible, reproducable comparisons of the broad range of DW/DSS solutions that are currently being proposed. > The only benchmark that matters is the one that models your system > and none of the TPC's model decision support yet. > > Neil Raden - nraden@ix.netcom.com In the final stages of a purchase decision, knowledge of a specific workload is certainly important. The product being selected is going to be running *your* workload, after all, not a benchmark. Of course, this applies to an in-house benchmark as much as to an industry benchmark. The real goal of industry standard workloads should be to reduce the range of exploration that must be done before the purchase is made. In-depth research into the strengths and weaknesses of a proposed solution can consume enormous amounts of time and resource -- any tool that allows you to get to a "short list" more quickly represents a valuable tool. I think TPC-D highlights the base capabilities necessary in a decision support system, providing just that sort of value. Obviously, I'm biased. I was a member of the committee that drafted TPC-D, and I work for a database company that provides tools for a full range of data manipulation problems -- OLTP and DSS. Take my (personal) oppinions with the appropriately sized grain of salt. jms ---- Jack Stephens I find that one worthless man is called a disgrace, Performance Analysis that two are called a law firm, Informix Software and that 3 or more become a congress! -- 1776