On Thu, 2003-06-26 at 12:00, Jonathan Bartlett wrote: > The spare CPU cycles only help if your bus can fill them. That's the > meaning of the phrase you quote. A 1Ghz processor is no faster and has > no more CPU cycles to spare than the 500Mhz processor (depending on the > bus speed - some newer buses go beyond this).
That was *NOT* the basis of the quote I gave. The basis of the quote was just as it appeared on the surface; folks buy bigger/faster CPU's because their friends trumped them and not because they need them. The bus speed argument also fails most of the time with modern CPU's with prefetch, cache, etc. Lots can happen to get data into the CPU and ready to process while the CPU is doing operations in the registers. Fact of the matter is, in the large majority of boxes the CPU system (including the bus) is running far below capacity the majority of the time. > In addition, you mention servers as being CPU-intensive. My own > experience has shown the desktop to be most CPU-intensive, while servers > are I/O intensive. That is so patently simplistic. It *depends* on the type of server and/or the type of desktop applications. Even on servers that one would think are I/O bound the box may actually be CPU bound if the wrong types of devices and controllers are used. I have seen NFS servers that are fully CPU bound; too many clients, too much space served, wrong type of controller and disk system. Web servers, esp. modern web servers running PHP or java servlets or the like, are almost always CPU bound. Mail servers with large client bases are often CPU rather than I/O bound. Do not make such dogmatic statements, esp. when there is such a large body of data to show that your generality is false in a large number of cases. - rick warner -- redhat-list mailing list unsubscribe mailto:[EMAIL PROTECTED] https://www.redhat.com/mailman/listinfo/redhat-list
