On Monday 10 October 2005 07:38 pm, Stefan Teleman wrote: > Agreed. For certain types of software, spec numbers are important. If > the software in question will spend most of its time doing what the > benchmark did. i don't know how realistic that is.
I don't know how realistic that is either, since it seems folks would be in the minority to do something specific like that. However, there are various tests and some do test more real world function, so I've been told. I've not run any of the SPECxxx tests myself, so can't comment on them to much, other than to understand what they test and how they test it. > Nope (if i am making a rational decision based on observable evidence > and facts). i deal with (soft) real-time software every single day of > my day job. this does not only include (near) real-time complex math, > it also includes (near) real-time network data tranfers (to the extent > real-time is possible). This is my experience as well, so I think we agree on this point also. People test for their environment, whatever that may be, and that doesn't mean the same thing as SPEC numbers in all cases. It is true that some of the SPEC tests do test some of the needed features for some folks, and this is why people follow them. > > I'd like to take a step back and concede that Linux is faster at some > > operations. I know Sun's performance group is working to tighten up that > > very gap in those areas. What does this all mean? > > sure it is faster on some operations, according to the benchmark. but > the benchmark is just one aspect, in my experience. Exactly my point. In the real world people need to get their job done, and that may or may not mean squeezing the last CPU cycle out of their systems. In fact, I argue that nobody wants to run their servers at such a high load, so many of the tests are invalid for the real world. Peak load is a different story, which nobody can plan for, and Solaris doesn't stay at peak load very long in my experience. -- Alan DuBoff - Sun Microsystems Solaris x86 Engineering _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org