Waaaay OT: OpenSolaris

[Bryan J. Smith <b.j.smith@ieee.org>]

At 02:10 PM 6/24/2005, Hoover, Tony wrote:
>Can someone tell me what advantages a "REAL" Unix kernel like OpenSolaris
>has over Linux?  Has anyone else thought of compiling OpenSolaris on Alpha?

From: Michael Huntingdon <hunting at ix.netcom.com>
> Why would anyone...especially in education consider this over Tru64 on Alpha?

Because Sun did a damn fine job.

I originally dismissed Solaris/x86-64.  I thought "why bother"?
But then I saw the designs.  And then the benchmarks in a recent review.

KEY POINTS ...

Linux grew up on the I/O-ignorant PC.
It assumes everything goes through the "front side bottleneck" of a
shared "memory controller _hub_."
E.g., All components contend for the Northbridge (MCH) ...
http://www.samag.com/documents/s=9408/sam0411b/0411b_f2.htm  

It only recently added NUMA processor affinity for programs and data.
But that still doesn't solve the I/O issue and the "hub":  
http://www.samag.com/documents/s=9408/sam0411b/0411b_f3.htm  

The AMD64 platform is the evolution of Alpha EV6 crossbar into an
EV6 NUMA + generic, partial mech of inter-CPU-I/O.  You now have
the reality where a kernel should optimally handle I/O -- including
"processor affinity" for memory mapped I/O and other transations.
http://www.samag.com/documents/s=9408/sam0411b/0411b_f5.htm  

So far, the Windows and Linux benchmarks have offered some
improved performance over Intel Xeon.  But not as much as I had
hoped.  Why?  Because Linux, like Windows, grew up on the "MCH"
PC platform.

Solaris, like Tru64, Irix and others, have grown up in NUMA and
partial-mesh CPU and/or I/O interconnects.  Opteron is the _first_
platform to make this commodity.

Solaris now runs on Opteron.  And damn if those benchmarks in
Linux Magazine wake me up to the fact that Linux, like Windows,
is severely lacking in "maturity" when it comes to optimally handling
such an system design.  But it makes sense, because Linux has
limited exposure to such designs -- even most non-PC\
implementations have only been uniprocessor or, in rare cases,
dual-processor.

And while it's still not GPL or GPL compatible, at least it _is_
under an MPL compatible license in most areas with even SCO's
blessing.  IBM AIX is not.  Tru64 is not.  In fact, as much as we
chastize Sun, they _have_ been at least "open" with things more
than IBM (although not HP, at least in its Linux endeavors, I'll
admit).

We have Sun to thank for LGPL'ing StarOffice, and several other
items.  Yes, their sticklers on Java, but even IBM's donations have
been largely _limited_ to Java under MPL (and non-GPL compatible)
licenses.  And Sun has _also_ been involved with many kernel
developments (like NFS), despite the lack of common attribution.

I'm not "advocating" Sun here, but don't knock Sun.  They are
doing their best in providing sound solutions.  Heck, SPARC is
at least an IEEE standard.  What's IA-64?  ;->


--
Bryan J. Smith   mailto:b.j.smith at ieee.org