Debian GNU/Hurd can now be installed a little easier. “This month Philip Charles created a new installation CD, the L series, for the Hurd, which brings us a big step towards installing the Hurd from the Hurd (without the need of a Linux-based installer). If you enjoy testing stuff, please give it a try.”
And maybe it will be ready for desktop usage in 300 years!
Damn, DNF will be old by then.
Its slow going but plz consider these guys are a small team. All the work is most probably done in their spare time.
So? Haiku is done by a small team in their spare time as well.
Even if you add the development time of the NewOS kernel (written by a single guy, back then), its development speed is still faster than Hurd’s.
Not by much.
Only if you don’t consider twice as long to be much.
LOL! Well, 10 years vs. 20 years sort of melds into “just a heck of a long time” to me.
Development of Haiku (as well as the NewOS kernel) begun in 2001.
OTOH, the first attempt to write a GNU kernel was in 1986. If you include the development time of the GNU userspace tools, it even started in 1984.
I could argue how Haiku even builds a complete desktop environment — something that’s totally out of scope for Hurd, but pictures say more than 1000 words.
So compare Hurd: http://commons.wikimedia.org/wiki/File:HURD_Live_CD.png
to Haiku: http://commons.wikimedia.org/wiki/File:Haiku_2008-02-19.png
First, I was a heavy BeOS user and I had Haiku running in a VM. Second, you CAN boot into X-Windows/Window Manager in Hurd, like so many of the old unix-like systems you have to set a flag for it to start up automatically.
Third, since I was a heavy BeOS user, and they got squashed by Apple and MS, I’ve been waiting for what seems an eternity for Haiku.
It’s all very subjective. As I tried to intimate in my previous post, 8,9,10,20 years all seem like a LOOOOONG time, my kids have gone from zygotes to relatively large speaking, reasoning creatures in the amount of time it has taken for an alpha version of Haiku to arrive. Do not misunderstand me, I hold great hope for Haiku (check my other posts on OS News) and am a big fan, and donate money to them.
I have maybe another 30 years left on this planet. I’d love to have a released version that I could do my actual work on as a daily OS. You brought up the Haiku comparison, I’m telling you to some people (ie. me) it is still a long time – no matter how you slice it.
Yeah and there are tons of other OSs written by a small team or one person, but you can’t compare them just by this.
Not to mention they difficulties added by using a microkernel architecture. It is doubtful whether the hurd will ever leave the lab stage, but it’s an interesting project, nonetheless.
think of the hurd project of a bacteria sleeping in a dry desert for some years until the rain:
If we all have a million processors in our devices to get some real artificial intelligence this is the time for that projects awakening. And it could have prevented some patents through prior art!
When did dead ends become interesting? Or do you mean that it is interesting to watch groups of people perseverate on dead end projects?
Edited 2009-11-11 15:20 UTC
The way I see things is
1. Finish the work on viengoos
2. Merge with Syllable
3. Release a Debian distro
Otherwise 300 years is a very optimistic estimate
This is great news, I have been waiting for a new release to test for some time. I will download and give the new L1 series a test drive.
Good, so finally the FSF will let Linux alone.
Just out of curiosity, have any basic performance tests been done against the kernel/operating system? How does hurd compare to linux or any of the BSD kernels?
Microkernel is necessarily slower than a monolithic kernel (unless the monolithic one is very badly done, of course).
OTOH, the cleaner architecture and modularity of microkernel affords for much faster development speed, as proven by hurd, minix and friends.
Did you manage to keep a straight face while typing that?
I did alright ;-).
NT, OSX, and QNX say “what’s up”
NT and OSX aren’t microkernels. E.g. read http://c2.com/cgi/wiki?MicroKernel (yes, there are more authoritary links, but who has time for that?).
QNX might be, though.
Both kernel teams for OSX and Windows consider their products microkernels, and operate and develop them as such. So I’ll take it from the fish’s mouth, rather than from someone who heard stories about fishing 😉
The BSD portion of OSX still is implemented as a server on top of mach. And same can be said about the NT kernel, even though there have been some modifications/deviations to try to make the common case fast.
Things evolve, but it would be the same as claiming how solaris is not an unix because it is not the very same as the original 1st edition of Unix 30+ years ago.
More interesting to me is how it compares to Darwin since they’re both using the Mach kernel.
It sounds a lot like Darwin, as mentioned by Avie Tevanian, in the fact that they readily tore the kernel apart and put it back together.
IMO, going with Mach was a very bad decision. Micro-kernels are necessarily slower, but Mach is just notorious for bad performance. They should have gone with something like L4 which is known for extremely good performance (for a micro-kernel anyways).
By the way, 5-10 years from now, performance of micro-kernels should be comparable to monolithic kernels. The reason micro-kernels are so slow is the constant context switching, which wouldn’t matter if the processor had a tagged TLB. With a tagged TLB, each TLB entry is tagged with an address space, so context switching would not require a full TLB flush, only a tag change (i.e. changing a register). Intel and AMD both have plans to introduce a tagged TLB in their processors sometime soon due to much better virtualization performance.
That should mean that Tru64 performed badly, right?
dude, processors have been using tagged TLBs (as well as other TLB managing techniques) for eons. Heck, I think the MIPS R4000 had it in silicon almost 2 decades ago. Commercial AMD64 procs have been using tagged TLB designs for a while too.
Furthermore, microkernels are not necessarily that bad when it comes to context switch issues (in fact context switch overhead hasn’t been an issue for the better part of the past decade at least due to the sheer difference between the quantum timer tick of modern processors and its pipeline clock, as well as things like out-of-order etc. Also, most of the servers in the microkernel are in the same privilege ring and don’t have to necessarily trigger a compound context switch.
A lot of people criticise X86, but it is funny how the baroque protection and privilege mechanisms, ended up being quite beneficial for kernel operation.
Many processor architectures such as MIPS have had tagged TLBs for a long time, but x86 has not. Intel and AMD have only recently started introducing tagged TLBs, any they are not fully used yet.
I think Xen and whatnot have been using/expecting tagged TLBs for a while now.
AMD64 has had tagged TLB literally from the get go. Pacifica and whatever it is Intel names their hypervisor technology have been around for a while too and expanded on that. I think you may be referring to things like nested page tables and such.
Further, x86 allowed for software management of certain TLB functions, as to make a more informed request for a TLB flush.
It is amazing that some people still continue to work on it. I first tried out Debian GNU/Hurd was around 2001. Since then, there have been some improvements (very little on the core os, mainly additional Debian packages being ported over to Hurd) But compared to anything else on the OS market, the speed is just so excruciatingly slow, that after following it for some years and occasionally running new builds on my computer, lost interest. I was taken by curiosity when I learned about the l4-hurd initiative to replace the dated mach kernel which its builtin drivers by something more modern, a real and efficient minimalistic microkernel, while drivers and almost anything else would really live in userspace. The design ideas have been fascinating, but the discussion on the mailing list (http://lists.gnu.org/mailman/listinfo/l4-hurd) went on for years and years. Markus Brinkmann actually wrote some initial code, but it never evolved beyond early embryonic stage, and finally, they decided that they need another microkernel. So back to start. If you look at the recent mailing list, they still seem to be engaged in design discussions, but no actual coding is happening. I guess the list has been around since approximately 2001, or maybe 2002. This is absolutely unique in the computing world.
Unfortunately, I believe that without a new kernel, Debian GNU/Hurd isn’t going anywhere. The old kernel hasn’t been developed by years, it doesn’t even have sound support. Let along Wifi, I think even USB isn’t in.
I have had strong sympathies for the project, but they probably never managed to attract sufficient manpower to get anywhere, maybe their style was too Cathedral-like, attempting to have the perfect plan in place before the first line of real code is actually written.
It is a loss, but I think the time for waiting for them is really over now…
Deciding on a new microkernel is key to the project moving forward. The developers identify this problem on their GNU Hurd Wiki. I wish they could settle on one and get the project moving faster.
I wonder how much of the current development work can be ported over to a new microkernel. My hope is allot of the work can be re-used, or else it will be like starting over again.
Another HURD related news item 😛
I really do like the idea, but I gave up years ago waiting for them to get anything serious running. It seems like every time they encounter any sort of problem, they spend a year or two researching the next great microkernel to base the thing on, and therefore actual development on the Hurd itself stays stagnant.
If you want to see the current state of the Hurd without relying on outdated experience and superstition of others, you can simply have a look at the official status page of the Hurd:
– http://www.gnu.org/software/hurd/hurd/status.html
It includes a screenshot of the Hurd running X with FVWM and Gnumeric.
And if you want external review, you can check a shot of GNUstep running on GNU/Hurd with a short status info:
– http://multixden.blogspot.com/2009/02/browser-on-hurd.html