SMP (Was: Why did you choose DragonFly?)

[Przemysław Pawełczyk]

On Mon, 20 Sep 2010 16:21:19 -0600
"Samuel J. Greear" <sjg at evilcode.net> wrote:


> (...)
> I will try to answer here and now.
> 
> The purpose of my question(s) is because I believe DragonFly BSD is
> not adequately represented in the easily accessible (main page and
> pages directly linked from it) literature on our website. I would like
> to fix this so that questions like yours need not arise in the future.
> As a first step I wanted to learn what people like about DragonFly and
> what keeps people using this OS and participating in the community so
> that I could expand on those points.

Hi,

Let me suggest you to find the Least Common Denominator from our
remarks, then "repackage" them and make small polls which could
be placed on common BSD websites and:

1) http://www.osnews.com/
2) http://distrowatch.com/

Why there? Because people comes from different worlds and even if they
resigned to use or to study DragonFly deeper they were attracted
by DF innovation techniques. It would be worth to ask them for your
review.

> DragonFly BSD is not necessarily targeted -to- SMP, rather it supports
> SMP and adopts a slightly different model than other mainstream
> operating systems. Most OS kernels use primarily "hard" locks in the
> form of mutexes and spinlocks. DragonFly prefers "soft" locks in the
> form of tokens and, even better, lockless algorithms. I do not believe
> anyone is ready to make any bold claims about our current SMP
> performance. However, I do believe many developers would claim that
> even though we may be a bit behind in performance today our
> model/methodology positions us well for the future.

Perhaps mere performance numbers do not apply here but other
interesting issues concerning SMP are strong or unique part of
DragonFly?

BTW
http://www.dragonflybsd.org/goals/

The page contains highly technical language. Perhaps two tier features
- for common BSD and Linux users and for developers - would be found
advisable? Most common users, those more advanced, will reach for the
developers' pages too at the end.

> Modern hardware should be pretty well supported, if FreeBSD supports
> it then generally so should we and if we don't it is likely not too
> much work to port that support over. Matt Dillon just brought up
> DragonFly on a new AMD 880-series-based motherboard with a 6-core CPU
> and that is apparently working quite well.

Most modern hardware (read as MoBo) comes as sort of chip families.
Most cheap boards are equipped with everything a common user needs to
run the modern operating systems - CPU, GPU, Gb Ethernet, USB, and
Audio.

Why not to point to the fact, if DF supports the families? In most
part the developers/contributors to other operating systems provide
hardware compatibility list based on every manufactured motherboard. It
was wrong and it is wrong in my view.

Example? Here you are. PC-BSD supports Nvidia chips, what makes that
some potential users back off reading the NVidia demand. But latest
Xorg drivers support new (2009) embedded ATI chips like ATI Radeon HD
4200 found in 785G Chipset. There are dozens of MoBos based on the
chipset family. There is out of touch of reality approach to
list every possible mobo - it make sense to name the family, bah, to
emphesize that we supposrt the whole families of chipsets (giving one
or two examples what do we mean by "the families").

> HAMMER is definitely a competitor to ZFS, although not in terms of
> volume management. We have LVM for that, but currently it does not
> support many of the targets one might expect (None of the real RAID
> levels). HAMMER is partly MPSAFE and many of the subsystems above it
> are also partially or fully MPSAFE, performance on multiple CPU's
> should be quite good from the point of view of the filesystem.

HAMMER is being seen as completely new system/invention. Why not create
a model configuration for Intel and AMD CPUs representative to common
user, check it and show how every subsystem of DF runs on the hardware?
If you say that AMD 880 works great, the same approach should be
shown with HAMMER on one or two disks.

In other words - show an example first, only then the users will be more
actively searching for existing documentation (e.g. manuals). As I know
no one from casual users starts with MANs and then installs system!
Just the contrary, the system should work, only then we reach for help.

If DragonFly is so invention laden then the prospective users should be
acquainted with working examples of the goodies. If not "should" then
decidedly would be to the "advantage".

Regards

-- 
Przemysław Pawełczyk (P2O2) [pron. Pshemislav Paveltchick]
http://pp.blast.pl, pp_o2 at o2.pl
Attachment:
pgp00004.pgp
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgp00004.pgp
Type: application/octet-stream
Size: 189 bytes
Desc: "Description: PGP signature"
URL: <http://lists.dragonflybsd.org/pipermail/users/attachments/20100921/05f765ab/attachment-0021.obj>