New != Faster
Kris Kennaway
kris at obsecurity.org
Mon Jun 4 18:16:05 UTC 2007
On Mon, Jun 04, 2007 at 12:54:18PM -0500, Tim Daneliuk wrote:
> In the course of trying to work through some problems with a new MOBO,
> I did some speed test which I found sort of surprising:
>
> Old System
> ----------
>
> Dual PIII 600Mhz w/768K Mem and Mylex RAID 5 with old 9G SCSI drived
> FBSD 4.11-Stable
> Writing a 1G file to /dev/null with dd reports about 26MB/sec
>
> New System
> ----------
>
> Pentium D 3.2GHz w/2G Mem and SATA Drive reported running at SATA-150
> FBSD 6.2-STABLE
> Writing a 2G file to /dev/null with dd reports about 50MB/sec
>
>
> So ... the new system should be much faster all the way around, right?
> Hmmmm, not necessarily so. 'buildworld' is only about 17% faster on the
> new machine v. the old. I would think that with way faster processors
> and twice the disk bandwidth I would have seen far reduced buildworld
> times. So, I decided to check a known fast machine. The results:
>
> Procs Mem dd Read OS
> buildworld
>
>
> Old 2 PIII @600Mhz 768K 26M/sec 4.11-stable/SMP 50-60
> min
> New Pent D (2 core)@3.2GHz 2G 50M/sec 6.2-stable/SMP 40-50
> min
> Fast 2 Xeon @3GHz 3G 130M/sec 4.11-stable/SMP 8
> min
>
>
> So, now I'm confused. These are all lightly loaded systems but the
> buildworld time does not scale even approximately by either CPU or
> I/O performance. What the heck is going on, I wonder? It is possible,
> I suppose that the "New" machine does not have SMP running properly on it,
> though 'top' shows two CPUs working away. Is the difference in speed
> attributable to 4.11 being faster than 6.2? Unfortunately, I cannot
> get 4.11 to boot on the "New" machine - it does not like the hardware
> for some reason claiming:
>
> RTC BIOS diagnostic error 80<clock battery>
>
> Even after I change the RTC battery on the mobo.
>
> Strange ... any input appreciated.
This comparison is 100% bogus.
4.11 and 6.2 are vastly different (the latter builds all sorts of
different code, and uses a *different compiler* that is slower in
compiling the code). When trying to compare something, you have to
compare the *same* thing, or it's meaningless.
Kris
More information about the freebsd-questions
mailing list