I was hanging out in San Francisco airport yesterday on a flight delay and went into one of the magazine stores to check out the latest monthlies. A bunch of them had the same full page ad from Oracle comparing a Sun server hosted Oracle DB with apps to an anonymous IBM “server” hosted Oracle DB with apps touting the TPC benchmark showing how the former blew away the latter in terms of performance.
I find such comparisons very amusing. For one thing, TPC benchmarks are hardly a valid comparator of anything. I know guys in Silly-con Valley who make their livings from configuring systems to win the TPC benchmark. They will be the first to tell you that they do not recommend the shortcuts that they take or the resulting configurations for actual systems working with actual company data. This is strictly NASCAR stuff: showing how cars can drive really fast in circles whether to the left or to the right. Has nothing to do with how the same car would do in actual traffic situations.
The box shown from IBM is a System z mainframe. I guess if you wanted to configure a z to do only one thing, you could probably get performance rivaling the x86 boys. But that would violate the whole architectural design and purpose of a mainframe. I find myself trying to explain this whenever I talk to a CIO who is still thinking about migrating mainframe workload onto x86 tinker toys, especially as a virtualization platform. The processor, memory and resource architecture of a mainframe is designed for sharing and scaling — which it does linearly, predictably and with enormous fault tolerance. x86, despite additional lines of code added for multi-tenancy, is not optimized for such purposes. Stack a bunch of VMs on x86 and you have a Jenga game: one app fails, all guests fail. Stack a bunch of apps in an x86 and you have lower and lower performance. At a minimum, I/O congestion is inevitable. The mainframe is not hampered by these things, as a function of its core architecture.
Can we please get some real comparisons and some real metrics to help guide our platform decisions? We need a Consumer Reports in tech, and have for a long long time. Brand name and marketing budgets do not substitute for architecture.
{ 2 comments… read them below or add one }
I disagree Jon – I think that ad is one of THE BEST ads in tech right now. Why? Because it forces vendors to engage in real debate about the performance of their hardware. End-users can decide for themselves which arguments are valid or invalid.
You are 100% right about the validity of the TPC benchmarks but the point is that it’s starts a discussion around performance and away from “which vendor has the coolest magic tricks”.
The misleading aspect is the apples to oranges comparison.
You must log in to post a comment.