NetBSD-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re:



I think that on some architectures there needs to be a more direct way of
loading without having to boot into the firmware. Right now, it's ppc32 and
soon sparc64. The former should have an  hfs bootloader you can dd to hard
disk. Two disks on a G4 and Net still needs me to change the default boot
device and type in the boot command.
I agree that BSD systems are cleaner and the performance on embedded devices
should be argument enough; however, popular opinion and business has different
ideas. Performance o my home machines is anywhere from 200 to 400% better with
any BSD than Linux.
The chewiefs looks like a good idea. I'm more of a Free, Open, Net BSD user in
that order.
--- On Thu, 11/25/10, andre.dolenc%kolumbus.fi@localhost 
<andre.dolenc%kolumbus.fi@localhost>
wrote:

From: andre.dolenc%kolumbus.fi@localhost <andre.dolenc%kolumbus.fi@localhost>
Subject:
To: netbsd-users%netbsd.org@localhost
Date: Thursday, November 25, 2010, 10:21 AM

Hello,

(Let me try posting again...)

+ Automated *performance* testing incl. kernel RAM usage, real-time and SMP
characteristics on given HW (e.g. latencies). Such testing will allow anyone
to check if, say, a new version of nbsd or a proposed change to nbsd improves
performance on own HW.
By way of example, take a look at http://www.inf.u-szeged.hu/csibe/ and
http://webkit.sed.hu/benchmark While the GCC test environment is open, and can
be downloaded by anyone, the webkit test environment cannot, unfortunately.

The work on-going in nbsd in developing ATS and related test cases is
absolutely -great-! Keep it up!

+ Anything that helps when porting nbsd to new HW, like kernel
instrumentation, tracing, and debugging.

+ Anything UI and graphics related (e.g., how do integrate the UI generated by
3D graphics and overlay on video output. Think of a camera UI for mobile: the
UI is transparent and overlayed on the viewfinder which is, essentially, a
video stream before the picture is taken); Wayland kind of window system,
etc.
Even if wayland idea is good for mobile, I am not sure the design and
implementation is the best possible.

+ BSD licensed flash file system. The folks at Szeged are slowing getting
there. As an introduction, these are the same folks that developed the current
flash file system in Linux, and before that they improved JFFS2.
http://chewiefs.sed.hu/

+ Performance improvements are always welcome, esp. related to code size,
configurability, (soft) real-time, SMP, and so forth. Things like a high
resolution user space timer come handy when doing some (mobile) audio stuff in
userland. Not to mention it comes handy when testing performance, like RT
characteristics on given HW.

+ Memory management can always get better. For instance:
- in some devices, it is possible to maintain a list of processes that can be
killed. This comes handy when the kernel runs out of memory. So, having a way
to provide the kernel with such a list is really good.
- Garbagge Collection and paging defeat each other. There is an article in the
ACM "Garbage collection without Paging" (main author is Mathew Hertz) showing
one way the VM can cooperate with programming languages that rely on GC (e.g.,
Java, Lua, Javascript, etc). Of course, such issues go unnoticed on servers or
desktops. It is possible that the changes done for Linux to support Android
include such improvements.

+ Instant-on. The work done to reduce Chrome OS boot time makes a lot of
sense, IMHO. Of course, many need to have more control over the boot process,
but end users don't care. Also, instant-on after sleep is great.

+ XiP support. New memory technologies could make nbsd extremely attractive if
and when it supports this.

+ Power management. Not much missing except when ACPI is not an option. A
clock framework is very handy when ACPI is not used.

In general, nbsd is a great kernel. Below, a few comments about the original
email:

    + please drop CVS,RCS,SVN . ...
We used git in our past projects, and we had no problems scripting our way
back and forth.

    + Support for more Filesystems. Whats going on in LINUX is crazy but ..
It is possible to use Linux filesystems on rumps, but that work never reached
completion. Unfortunately. Incidently, the feasibility study was done using
JFFS2.

    + NetBSD LIVE CD/USB with each release ...
    + integrating a small X-Window into the System. Something like Project
"Wayland" from Canonical that has a minimalistic mem footprint expandable by
userland modules.
IMHO, take and improve the wayland ideas, and make it possible for others to
use X. But don't let X compatibility get in the way of a good solution. Let X
folks deal with the problem, if one arises.

    + NUMA support and a true scheduler fÃr each MultiCore CPU
    + More focus on Mobile Devices - NetBSD is in my opinion a better System
on mobiles than LINUX/Android. Tethering drivers (to iPhone and Android) would
be nice.

I definitely support the focus. I also agree on your opinion (at least in the
Android context) even though its unlikely we would agree on the reasons.
"Better" is in the eyes of the beholder.

Thank you for your time.

Disclaimer: the opinions above are my own and don't necessarily reflect those
of my employer, colleagues or anyone else. I present these opinions solely as
an individual.


Home | Main Index | Thread Index | Old Index