I said:
>For a very interesing comparison of Solaris 2.4, FreeBSD 2.0.5R, and
>Linux 1.2.8 on the exact same hardware (P100), see
...
>You'll have to make the assumption that the relevant architectural details of Solaris,
>and the resulting strengths/weaknesses, are consistent between the Sparc and
>Intel implementations.
On futher reflection, that's perhaps not a totally valid assumption.
One of the main conclusions of the paper cited was that none of these
OSes properly takes proper advantage of the Pentium's architecture for memory
writes (presumably to be compatible with lesser x86's):
Our results show that none of the systems adequately delivers the
Pentium's memory write performance. For example, the Pentium can
copy data at over 160 megabytes/second using a prefetching copy
routine, yet none of the systems we tested have implemented such a
routine. As described below, the prefetching routines address the
fact that the Pentium does not have a write-allocate cache.
Without this optimization, the same routines copy data at about 40
megabytes/second.
I'd expect that a Sun OS running on a Sun box would take full advantage
of the hardware's capability. I cannot parlay any of this into hard
figures for your evaluation, though. The tests used memcpy and memset,
figuring the kernel used the same, or similar, routines. (They didn't
examine any kernel code.) I wouldn't be too surprised to see any or
all of these OSes in future versions adapt to the type of hardware, at
least in the kernel, if not in the std. C library routines.
This very interesting (to me, at leaast) subject risks getting too far
afield from the proper topic for this list. I humbly suggest that
anyone who wants to delve into these OS/hardware issues to a degree not
related to firewalls do so in a more appropriate forum.
- KH
|
|
