Barrelfish Operating System | The Coffeehouse

TommyCarlier
I want my scalps!

Cool new research OS from Microsoft Research: Barrelfish. Source code available under BSD-like license.

Barrelfish is a new research operating system being built from scratch in a collaboration between ETH Zurich and Microsoft Research, Cambridge. We are exploring how to structure an OS for future multi- and many-core systems. We are motivated by two closely related trends in hardware design: first, the rapidly growing number of cores, which leads to a scalability challenge, and second, the increasing diversity in computer hardware, requiring the OS to manage and exploit heterogeneous hardware resources.

Dr Herbie
Horses for courses

I like the page footer: "This web page was brought to you by a server running Barrelfish.". Smiley

Herbie

Bass
www.spreadfirefox.com/5years/

I compiled and ran it. Very strange OS. The shell is very b0rked. I'll post a SS if you want, but there is not much to see. Smiley

Charles
Welcome Change

In reply to Bass

#Sep 28th @ 10:48 AM

The interesting part of this OS has little to do with the shell at this point.... Smiley
C

exoteric
I : Next<I>

In reply to Charles

#Sep 28th @ 11:15 AM

So now we have the following

Windows 8, 9, ... client/server research/development
Singularity research/development
Barrelfish research/development
Azure research/development (David Cutler et al)
Midori (Jonathan Shapiro et al?) (codename for?) (dead/alive?) (incubation/next version?)
unknowns

Interesting times.

I still think some of the most interesting things are, deciding implementation languages. And of course isolation concepts, security, virtualization etc.

It's an interesting jigsaw puzzle to follow.

magicalclick

In reply to Dr Herbie

#Sep 28th @ 4:44 AM

Ditto, it works

ManipUni
Proving QQ for 5 years!

While this OS sounds interesting I'm still more interested in Singularity. Just seems like what the future will hold with or without Microsoft on the train. With machines being relatively cheap and human's time being relatively expensive managed code's benefits are very strongly felt.

magicalclick

In reply to ManipUni

#Sep 28th @ 12:53 PM

Well, I think those are not as cool as super cloud OS. Imagine the OS holds no user data or apps, just a bare bone connection to the physical HW. Then, you load the VHD that contains all the setup about security, registry, envirnment variouables, and etc. But, not everything is in the VHD, when you run the registered program, it will then download it from cloud (should use .Net for small footprint). You can see your files, but, only downloaded when you open it.

So, it is like a roaming profile without really download everything at once, only download by demand. Only the VHD is downloaded at once with all the OS properties and registry and etc to be loaded. And it does cache everything for offline use. Now, that the real OS on the go. I could have my notebook, netbook, desktop, and everything runs in the exact same VHD. ofc I will still need to buy the bare bone OS to read that VHD.

Pretty much a glorified Mesh.

PerfectPhase
"This is not war, this is pest control!" - Dalek to Cyberman

In reply to magicalclick

#Sep 28th @ 1:13 PM

On the app download on demand side of things, you had a look at App-V yet.

Our terminal server pool is going in the direction that you suggest. Hyper-V R2 clusters providing a standard HW layer, a standard 2008 R2 RDS image with no software, all software loaded on demand from an App-V server. Need more capacity, deploy a new image. No need to update the base images every time you update an app, or have to wait for SCCM to deploy the apps etc to each node.

The nice thing about the App-V technology is it only downloads the blocks (dll,exe etc) that you need for the features you use. So if I build an image for say Office 2007 Enterprise, and deploy it to a laptop user, and they only use word, they'll only pull the data required to run word, not office, excel et al.

Would be really cool if this could start to become the defacto install technology rather than msi.

magicalclick

In reply to PerfectPhase

#Sep 28th @ 3:05 PM

Sounds really really nice. I hope that would be the next gen Windows Live. =) Upload my own image and download on demand. That's true Windows Live.

Bass
www.spreadfirefox.com/5years/

This downloads and builds it on Ubuntu 9.04:

--------------------

sudo apt-get install ghc6 libghc6-parsec-dev libghc6-parsec-prof build-essential cmake latex2html qemu graphviz

wget http://www.barrelfish.org/barrelfish-20090914.tar.bz2

tar xjf barrelfish-20090914.tar.bz2

cd barrelfish

mkdir build

cd build
cmake ..
make

--------------------

Run it with:

-------------------

make sim

-------------------

This will open the Qemu virtual machine and boot into Barrelfish.

If someone wants to post Windows instructions feel free to. Smiley

Bass
www.spreadfirefox.com/5years/

It's a pretty interesting codebase by the way. Essentially it's running a new kernel for each core. Smiley

staceyw
Before C# there was darkness...

Why another OS instead of forking and using Singularity as a test bed? That is was one of the reasons for it anyway, to allow research and testing new ideas on it.

Bass
www.spreadfirefox.com/5years/

In reply to staceyw

#Sep 28th @ 8:29 PM

Singularity hasn't had a single update (at least publicly) since Nov 14 2008.

exoteric
I : Next<I>

In reply to staceyw

#Sep 28th @ 8:29 PM

A specific set of ideas and languages. In particularly writing almost everything in Sing#. You're implying everything else can be built on these ideas or use the same language. This is obviously not true. It looks like Microsoft are spawning several "small" OS projects to investigate some concepts and see how/if they work in practice. It makes sense.

Creating a new OS is such massive investment and you want to be pretty damned sure you don't go down a path that is not sufficiently strategically safe, meaning you have not just picked the "first and best" ideas you came upon but really tried out several radical ideas before settling on something.

It'll have to live for decades going forward. I remember WPF being called a strategic API with an expected lifespan of at least 15 years. Now imagine an operating system. Just creating it means it'll have to live for a long time as well. Although hopefully it will be written in such a modular way with such language(s) that code can be reused in a possible future split when the next paradigm shift arrives; quantum computing or who knows what.

staceyw
Before C# there was darkness...

In reply to exoteric

#Sep 28th @ 10:38 PM

I agree with that. But many of the new ideas in terms of concurrency can be tested without starting from the metal (i.e. message passing (ala Erlang), etc). Flush those out, and start with something else if needed. If singularity somehow failed or will fail in that regard, that would be good info also, as your one step closer to a solution. Maybe it is time for another summary on what has worked and what will not work.

Dr Herbie
Horses for courses

In reply to staceyw

#Sep 29th @ 5:59 AM

I suspect it's a matter of simplicity -- when you wish to investigate a specific aspect of a model, it's easier to build a new model from scratch so you don't have to work through noise from other aspects that were looked at in the past or from features that have no direct bearing on what you're interested in.

Herbie

Bass
www.spreadfirefox.com/5years/

Has anyone besides myself actually tried this OS?

CannotResolveSymbol
{insert caption here}

In reply to Bass

#Sep 29th @ 7:31 PM

I did. Interesting to see it boot, but there's not that much there yet beyond that.

I'll be keeping a close eye on it, though. Seems like it has promise (even just as a research OS).

(I really want to stuff it in /boot and see if I can load it with Grub on real hardware... not sure if it would actually work with my hardware, though.)

contextfree

In reply to exoteric

#Sep 28th @ 12:42 PM

Would it be possible to add a SIP mechanism to an existing OS only for Sing# (or some other form of statically-verifiably-isolated) code, with unmanaged code defaulting to standard protection mechanisms? You could even support multiple SIP systems for multiple languages/runtimes ... is there some reason this couldn't be made into a feature of an existing OS? (would interop be problematic?)

Bass
www.spreadfirefox.com/5years/

In reply to CannotResolveSymbol

#Sep 29th @ 7:43 PM

Apperently the researchers did test it on real hardware, but I assume if you don't have basically the same exact model of PC they tested it on, you'll be SOL. Smiley

Bass
www.spreadfirefox.com/5years/

In reply to contextfree

#Sep 29th @ 9:06 PM

I don't know how it is in Windows, but in Linux+Mono I am pretty sure all Mono apps run in the same address space.