Stephen Hahn wrote:
>
> * I know I tend to be on the "let's change things" side of development
> here, but I am a little surprised by the assumptions that got us to
> this point in the discussion. At no point have I heard an actual
> argument why OpenSolaris's kernel and core libraries and utilities
> should target a small, lightweight platform. Such a choice has an
> impact on architecture and development practices; this impact has a
> cost. I've seen no suggestion that the benefit would outweigh those
> costs, or even what the benefit might be.
>
I'm to blame for the thread taking this slant I think.
Honestly, I've wished numerous times that I could use OpenSolaris for
certain projects. One of these we are stuck with Linux because:
a) NetBSD's pthread implementation is busted
b) OpenSolaris is too hard to minimize adequately
c) Perhaps FreeBSD could be used, but there were other concerns
I hate using Linux, and I hate being stuck with the GPL. The CDDL is a
much more attractive license.
I've proposed that OpenSolaris/ZFS/NFSv4 are the killer application for
a consumer-grade NAS appliance. This appliance would run from a fairly
tiny CF device, using ZFS and SATA drives for storing _data_ (as opposed
to the OS image.) The reason to use a tiny CF is to minimize power and
space footprint, and mount it read-only with /var copied into tmpfs to
minimize damage that can be done via hacking or intrusion.
ZFS has some clear, huge avantages over Linux in this space, and
Solaris' MT support can't be beat (use a Core Duo).
So that is the concrete rationale.
Then there is the general feeling that operating systems in general (and
Solaris in particular) have suffered from a general cavalier attitude
about the availability of system resources from developers. I have
often heard the refrain that "don't worry about an extra
cycles/megabytes (or ten, or even hundred), by the time we release
nobody will be running systems where this will matter."
That attitude sticks in my craw. Just the waste heat generated globally
by systems spending cycles doing extra swapping boggles the mind. Never
mind that instead of systems being faster and zippier, the software we
generate today tends to be bigger and slower. I still have to restart
my X session once a week because various X applications have sucked all
the memory out of my system (firefox/flash are the biggest, but not the
sole, culprits.) Why?
Never mind trying to scale a system to support hundreds or thousands of
Sun Ray users. It doesn't scale well, because developers of
applications (desktop applications are amongst the worst) just assume
that everyone has their own multi-gigabyte machine and that it isn't
worth their effort to tune their code.
I'm all for developing new features. But I'm also all for going through
old code, cleaning up unused or stale code paths, optimizing things to
run better, and eliminating redundancy. I'm even more for figuring out
how to do more with smaller more efficient CPUs (my primary NFS server
runs on a VIA C7 that consumes a meager 25W) without having to sacrifice
core functionality.
-- Garrett
_______________________________________________
opensolaris-code mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
