ideas to opensolaris as desktop

[kraftman]

yeap maybe is ath5k driver the other dont recognize my card, about blobs well i still prefer a blob to play in my laptop that going to windows ofc oss driver for nvidia at least would be peachy, aka blob until noveau grow more

If I wouldn't have Windows installed and if I would have a choice to use blob or OS driver right now I would use blob to have great 3D support It seems OS drivers have better 2D support.

About Madwifi vs Ath5k it's the first time when Ath5k works better for me (it started working with Kubuntu 9.10-rc) and this is great, because Ath5k is in kernel, so I don't have to care about compatibility. While there's no stable API in Linux the best option is to have drivers in mainline IMHO.

[jrch2k8]

Afaik Linux is the first with USB3 support:

http://www.neowin.net/news/main/09/0...support-usb-30

yeap maybe is ath5k driver the other dont recognize my card, about blobs well i still prefer a blob to play in my laptop that going to windows ofc oss driver for nvidia at least would be peachy, aka blob until noveau grow more

[kraftman]

Hmm ... I cant find the USB3 thingamy anywhere, I saw a page on the screen of the ex-Sun guy in our office at the time it happened, which i believe was from a guy internal to Sun at the time (before i formatted his laptop and installed Ubuntu on it) ...

I'll retract it as uncited, fair play and all

Afaik Linux is the first with USB3 support:

http://www.neowin.net/news/main/09/06/11/linux-is-first-os-to-support-usb-30

[kraftman]

thinking about it airlie is rigth on some points, but i think what is making the big diff about driver stability is the nature of both kernels aka in linux if you want stability you cannot upgrade your kernel or distro excepting maybe enterprise distro like rh or novell cuz drivers in linux are too tied to the kernel inner api. for example my atheros card worked fine with kernel 2.6.30.6 but something break wpa2 auth in 2.6.30 .8(i think was 8) later 2.6.31.2 auth worked fine but it keep getting disconnected randomly then in 2.6.31.11 wpa2 auth is broken again aka never auth, this is my karmic office laptop. in my toshiba x205 home game laptop in 2.6.30 series nvidia driver worked relatively fine but it got a flicker that i didint have with jaunty 2.6.28(same driver version ofc), now in 2.6.31 the drivers never work unless kernel boot with acpi=off(seems that libc break something in acpi userland or backwards) but acpi=off broke initd script to load swap, so upgrade your kernel is all about luck any revision can break stuff in the nastiest way or it can work beautyfully. on the other hand in opensolaris i made a full upgrade from 200906 to 2010xx and so far nothing, even nvidia drivers are working without touch anything so i assume that opensolaris handle drivers with some sort of more stable outer api so it feels more stable

Actually I have an Atheros based card. I compiled madwifi driver and after upgrade from 2.6.31.3 to 2.6.31.4 driver is working. However, maybe because of things you described they're making ath5k . Binary blobs don't interest me at all If everything is about stable vs unstable API I don't see much sense in continuing this, because some people prefer stable and some other people don't. There's also possibility there aren't many changes in Solaris kernel and thus it doesn't break things. I only 'tolerate' drivers which are provided with the Linux kernel archive - they shouldn't break too often if ever () and if I'm saying they're stable etc. I'm only according to such drivers.

[jrch2k8]

thereis any chance technically speaking to replace drivers inside the kernel for usermode drivers??
for example

* remove all drivers from kernel, except stuff needed like cpu, pci, etc
* make a set of kernel modules that provide an abi stable basic functions to access hardware keeping security ofc
* make a set of library with an stable abi (maintained for kernel.org or someone close). this library should have all the routines to access the hardware like libpci for example.
* drivers can be .so files or something similar

i think this way kernel dev would be faster, cuz you only have to worry about implementing or fixes real kernel issues and and not hunt 3k broken drivers cuz someone changed a function name, kernel maintaniners only have to adjust this changes into this kernel modules and library or even add new one but keeping it abi compatible so drivers can keep working and when updated drivers can use the new goodies

i think something like this will bring the only thing still missing in linux kernel, driver stability through releases and ofc if maintainers use their time polishing the kernel instead of hunt broken drivers, well i think good think will come too

and ofc more drivers stability, with this commercial drivers focus on fix the driver and not to rewrite every time every kernel is out

[jrch2k8]

USB3 on Linux support is already upstream in the kernel, I've no idea where you heard that story [citation needed], but Intel has put nearly all their development work on USB3 into Linux so far in public.

Also I'd be surprised if you can find any 10GBe cards that are better supported in Solaris than Linux.

Solaris has been on a downward trend since they dropped x86 support that time briefly and really I can't see it coming back. I understand why Solaris drivers feel more finished because there are less of them and the ones that actually get into Solaris releases would have passed a lot of QA. However my guess is if you buy hw to run Linux on as carefully as you have to buy hw to run Solaris on, you'll most likely get a lot of parity in the driver support. I know how much work goes into the RHEL ethernet and storage drivers, and from what I can tell this is substatially more than goes into Solaris equivs at this time (5-10 years ago I'd say quite the opposite was true).

DAve.

Dave.

thinking about it airlie is rigth on some points, but i think what is making the big diff about driver stability is the nature of both kernels aka in linux if you want stability you cannot upgrade your kernel or distro excepting maybe enterprise distro like rh or novell cuz drivers in linux are too tied to the kernel inner api. for example my atheros card worked fine with kernel 2.6.30.6 but something break wpa2 auth in 2.6.30 .8(i think was 8) later 2.6.31.2 auth worked fine but it keep getting disconnected randomly then in 2.6.31.11 wpa2 auth is broken again aka never auth, this is my karmic office laptop. in my toshiba x205 home game laptop in 2.6.30 series nvidia driver worked relatively fine but it got a flicker that i didint have with jaunty 2.6.28(same driver version ofc), now in 2.6.31 the drivers never work unless kernel boot with acpi=off(seems that libc break something in acpi userland or backwards) but acpi=off broke initd script to load swap, so upgrade your kernel is all about luck any revision can break stuff in the nastiest way or it can work beautyfully. on the other hand in opensolaris i made a full upgrade from 200906 to 2010xx and so far nothing, even nvidia drivers are working without touch anything so i assume that opensolaris handle drivers with some sort of more stable outer api so it feels more stable

[jadrevenge]

Hmm ... I cant find the USB3 thingamy anywhere, I saw a page on the screen of the ex-Sun guy in our office at the time it happened, which i believe was from a guy internal to Sun at the time (before i formatted his laptop and installed Ubuntu on it) ...

I'll retract it as uncited, fair play and all

[jadrevenge]

I'll agree to differ in my opinion, if you will.

Jon

[airlied]

For instance the Networking system in Solaris that enables 10Gb ethernet (http://opensolaris.org/os/project/crossbow/) has been significantly modified to enable full speed CPU utilising streaming of network traffic ... It's the first system to get that kinda performance ... and the Manufacturers were pretty much lining up to make sure that their cards worked with it, and were the best they could be.

With USB3 (not yet released) the first box in the world to run it was a Solaris box with a modified USB driver at Intel ...

My point was only that the drivers on Solaris have a more finished feel to them on the whole, and as you've probably noticed I have a tendency to waffle.

USB3 on Linux support is already upstream in the kernel, I've no idea where you heard that story [citation needed], but Intel has put nearly all their development work on USB3 into Linux so far in public.

Also I'd be surprised if you can find any 10GBe cards that are better supported in Solaris than Linux.

Solaris has been on a downward trend since they dropped x86 support that time briefly and really I can't see it coming back. I understand why Solaris drivers feel more finished because there are less of them and the ones that actually get into Solaris releases would have passed a lot of QA. However my guess is if you buy hw to run Linux on as carefully as you have to buy hw to run Solaris on, you'll most likely get a lot of parity in the driver support. I know how much work goes into the RHEL ethernet and storage drivers, and from what I can tell this is substatially more than goes into Solaris equivs at this time (5-10 years ago I'd say quite the opposite was true).

DAve.

Dave.

[jadrevenge]

Sorry, I really shouldn't have posted that ... it was only going to be flame-bait.

I've written both Solaris drivers _and_ linux drivers.

If you look for me on the internet you'll find my name next to the Canon scanner driver under SANE and related kernel changes that i needed to do because the SCSI interface on Linux was written by a guy who wanted to get _his_ scanner working and didn't know what settings were useful for everyone else, just what worked for him. At one point in time they create a packet of information and send the packet and the size to another procedure where it throws away the size and passes it to another procedure that guesses what the size should have been based on the first element in the packet.

Sorry even after all these years I still feel slightly bitter ...

The Solaris drivers I wrote weren't great though, they were for a USB scanner, in Solaris 8, whilst the USB bus in Solaris 8 was new ... if you powered off the scanner (or got a paper jam) it halted the bus and you had to force a bus reset before it would work.

Look, where a manufacturer posts a driver, it in general has been through much more testing ... Check out the Intel drivers for linux.

Where a driver is written by the guys that are paid to keep the kernel of an operating system clean and working in conjunction with the manufacturer it makes sense that the driver is gonna be good.

For instance the Networking system in Solaris that enables 10Gb ethernet (http://opensolaris.org/os/project/crossbow/) has been significantly modified to enable full speed CPU utilising streaming of network traffic ... It's the first system to get that kinda performance ... and the Manufacturers were pretty much lining up to make sure that their cards worked with it, and were the best they could be.

With USB3 (not yet released) the first box in the world to run it was a Solaris box with a modified USB driver at Intel ...

Look don't get me wrong, I've had sh*tty drivers on Solaris, I have a machine sitting opposite me that has a Broadcom 5787 onboard card in it that works 1/8th the time, and that was during install after patching the install server and crossing my fingers, and in the past I've managed to completely foobar a machine by installing a schilling driver to access a CDrom drive on a machine with SCSI disks ...

When my new laptop arrives in 4 days (It comes with Vista <SHUDDER>) it will have Ubuntu on before I let my family near it so that it has proper 3d support, very useful if you wanna do silly little things like watch BBC iplayer with no lag, or play games on lego.com without thawing out an Iceberg with CPU usage, that I'd have to do if I installed Solaris on the thing ...

I'm possibly undervaluing Solaris at this point, but I truly wouldn't put my family through something on the desktop that feels unfinished at the moment, they're not exactly computer literate to administrator level, my eldest being only 8.

My point was only that the drivers on Solaris have a more finished feel to them on the whole, and as you've probably noticed I have a tendency to waffle.