>>>>> "jj" == Jaka JejÄiÄ <jaka%jejcic.com@localhost> writes:
jj> We have a niche OS that is good, actually it is the very
jj> best in its niche -> and expert user.
Except the niche keeps getting smaller each year. I used to think the
NetBSD niche was embedded systems, but now almost everyhing I can
actually buy runs OpenWRT instead. NetBSD only runs on some old
expensive boards you cannot get any more. OpenWRT has a
GNU-configure-based packaging system that can build everything with a
cross-compiler, and a flash-friendly filesystem albeit one that's only
suitable for very small flash like <0.5GB, and they have
execute-in-place on NOR FLASH. The build tools and flash fs work
together so the end result is an image you can burn into flash, and
once in flash, it's *WRITEABLE*, it's not just some lame ramdisk. and
the whole openwrt distro is maintained in relatively good order by an
extremely small team. and now there's this new sort of post-Unixy OS
in the google phone? As a netbsd advocate I feel pretty defeated.
It will be good to finally have working threads for the first time
ever when 5.0 comes out. that will really change everything, night
and day. to be honest I did not even bother building 4.0 anywhere,
and 5.0 I will definitely try again.
but.....I think ``expert'' ``user'' are sort of two separate words.
meaning, yes you are an expert, but all you do is use NetBSD, you
don't work on NetBSD? use it, for what? for running a
single-threaded C web server serving up static pages? tar and untar
things? how about:
* fancy programming languages. Java, Lisp, Haskell, Erlang, modula2.
Forget the postgres vs. mysql ranting. How is our CouchDB/Django
stack doing?
* fancy modern routing stuff. like policy routing and
equal-cost-multipath. Cisco has cheap devices running IOS using
less than 5 watts doing such things. OpenWRT can do them too.
* multi-terabyte filesystem with O(n) or better fsck
* fancy security stuff. NFSv4/ZFS/Windows-style ACL's. MAC labels
and ``policies'' like classification. we have only systrace, no?
* fancy cluster stuff. GFS, OCFS, GlusterFS, Lustre? support for
Infiniband or ethernet RNIC's? granted, even where this works it
sounds like it's buggy as all hell, but if NetBSD is a good place
for kernel developers you'd think cluster operators would be
leaping over each other to use it. since scientific cluster admins
run almost nothing but small self-written program rather than
mountans of Linux Desktopware or Web Crapplets, and there are so
few of them they all have to be kernel hackers. yet they are all
camping in the Linux tent. Maybe because all the tools they depend
upon were wrenched out of the grubby claws of the corproations who
wrote them through the GPL, and if anyone did start work on a BSD
cluster, their work died along with their masters' VC funding?
dunno, probably shouldn't speculate, and I'm sure the truth is
secret anyway. wouldn't want to embarass any potential supporters,
even after the fact, right?
* fancy virtualization stuff. we are pretty good here in NetBSD with
Xen dom0/domU when they are up-to-date, the only BSD to have them!
But, on Linux/Solaris you can choose Xen, VirtualBox, VMWare, UML,
jails/zones/foreign-branded-COMPAT-zones, even lpar/sun4v.
* fancy powersaving stuff and thorough support of native hardware.
suspend/resume, hibernation, CPU frequency scaling, hotplug of
disks, PCI cards, memory and processors, rewrite of ECC errors.
Other OS's actually do this. For example thinkpad X-series docking
station is handled as PCI hotplug under Linux!
* fancy email. yeah you can run Postfix and dovecot with NetBSD, but
the web2.0 kids are using these big hairy Email Overlord Packages
like Zimbra, Scalix, iPlanet that are such spaghetti programs
they'll probably never work well except on their native OS. but if
you do not offer something similar-sized, any reasonable user will
use Outlook or Gmail instead. It is a decade later now, and sorry
but dovecot no longer cuts it.
* fancy realtime scheduling. well, nobody _really_ has this. but is
NetBSD's kernel as fine-grained preemptable as Linux's at this
point? Are the fancytimers finally working? (I'm having some
problems with them on Linux I think.) How about dividing
interrupts among CPU's in an SMP system, or partitioning certain
jobs vertically with CPU affinity to avoid lock contention like
Solaris does in several spots? Linux is certanly well-flogged in
that department, and I'm not sure we've made any progress while
the other tent has.
In the end, can I really do things with NetBSD I cannot do with
Cygwin? or, is NetBSD just like ``cywin lite,'' faster to install
faster to run and less extra baloney than Cygwin? because, if that's
all, well then I have ``cygwin ultralight''---an NSLU2 or Fonera 2.0
Beta running openwrt---and those only use like 5 or 10 watts and
extremely simple. :) Their rc scripts are stunning, actually---very
NetBSD-ish, but *very* simple.
We're missing a lot of expert-user ``fancy'' things. A lot of that
list is probably fixed already since I've been a dropout for the whole
4.0 years, and haven't tried 5.0 yet. but it seems drowningly
overwhelming does it not? NetBSD is probably still okay for an expert
kernel hacker---not sure how the core-dumping, kernel debuggers, and
lock infrastructure are doing lately if NetBSD is still way ahead but
I doubt it could be behind Linux and that stuff is hard everywhere,
and good cleanly-written architecture is probably still the most
important thing. but, not the most important for an ``expert user''.
only most important for an expert developer. and it's so tempting, if
I'm interested in RDMA say, to start with an OS that already has
RNIC's and IB cards working, even if it's spaghettiOS.
Attachment:
pgpxYmwqG1X3g.pgp
Description: PGP signature