"Though most people don't know it, Microsoft has, not one, but two new operating systems that it is working on. The first, Vista, many people have heard of. Currently scheduled to ship just after the second coming of Christ, Vista has garnered most of the spotlight. However there is another OS lurking in the basement in Redmond and its name is Singularity (.pdf). So what does Singularity look like? A joke, at least at first glance. But taking a moment to analyze the situation I came up with some interesting observations which I will now share with you."
Post a Comment
Member since:
2005-07-06
Member since:
2005-12-02
"What they most definitely wouldn’t choose is Windows. An embedded system, just like a server, needs an OS to be stable and reliable, neither of which are areas that Windows is very good at."
Actually this is quite a funny statement. It is not the OS to blame for security, but the admins. Windows is just as stable if administered correctly. Stable does not have to do with rebooting after a patch, but whether it can handle the load placed against it. I am a sysadmin, and from my experience it is the lack of training for admins that makes ANY os insecure and unstable. I have not seen a "Blue Screen" since Windows 98SE. For servers I have not seen one just crash to crash since NT 4.0, unless it is administered poorly. For embedded, I have had to reset Palm OS devices more then Windows CE devices. Obviously the author lacks experience in what he is writing about.
Member since:2005-07-06
> It is not the OS to blame for security, but the admins.
> Windows is just as stable if administered correctly.
This is simply not true. In Windows most of the security is based on the infallability of programmers and code auditors. There are other OSes in which it's relatively easy to build very secure systems even if there are bugs in the software. And no, Singularity is not very good in this department either. Go check out what the EROS/CapROS/Coyotos people have and you'll see how security is done The Right Way(tm).
> [...] I am a sysadmin, and from my experience it is the lack
> of training for admins that makes ANY os insecure and unstable.
Just because a sysadmin can make any system insecure doesn't mean that some particular system can be made secure.
E.g., in Windows a program can do whatever the user running that program can do, and user privileges are very static by nature. Because of this, and because there will always be bugs in programs, Windows will never be very secure. And yes, most unices (OSX included) experience the same problem.
Member since:2005-10-25
Go check out what the EROS/CapROS/Coyotos people have and you'll see how security is done The Right Way(tm).
Ok, show me a computer running Coyotos. My laptop runs Fedora Core Linux. It's working now, well, today. Can you get Coyotos to boot? Run a program?
If by "secure" you mean "not even the owner can get it to do anything", Coyotos is indeed something remarkable. Otherwise, it's a research project that never made it to anything even remotely viable.
Member since:2005-07-06
> > Go check out what the EROS/CapROS/Coyotos people have and
> > you'll see how security is done The Right Way(tm).
>
> Ok, show me a computer running Coyotos. My laptop runs
> Fedora Core Linux. It's working now, well, today. Can you
> get Coyotos to boot? Run a program?
I'm trying to figure out your point... Since what you quoted was a response to windows being secure, then I guess that you're trying to say something like: "Windows is secure because your laptop can't run Coyotos." or "Coyotos is not secure because it's not complete."
That's of course utter BS, but I probably misunderstood what you were trying to say, so please correct me if this is the case.
> If by "secure" you mean "not even the owner can get it to do
> anything", Coyotos is indeed something remarkable.
That has nothing to do with security, and I didn't say anything along those lines. The state of a project is completely orthogonal to whether its designed to make exceptional levels of security easy to achieve or not. E.g. Windows is very mature (at least in that it has been around for a while), but still its various designs make it nearly impossible to make things secure (while not becoming unusable) on almost any level.
> Otherwise, it's a research project that never made it to
> anything even remotely viable.
"That never made it"??? Apparently you have no idea what you're talking about. Coyotos is a very new project. However, its predecessor, EROS, does indeed work, and CapROS is the continuation of EROS.
You seem to think that a low number of successful attacks is what defines a system as being secure. This implies that pretty much everything that isn't popular is very secure. That's of course BS, as vulnerabilities exist, or don't exist, regardless of whether someone manages to use them or not. The fact that you don't necessarily need security if your system is obscure or uninteresting enough is a completely different thing.
Member since:
2005-07-06
Man, this reads like a bad Slashdot post. Well, might as well reply to it like it were one:
Though most people don’t know it, Microsoft has, not one, but two new operating systems that it is working on.
Microsoft per se isn't working on Singularity, Microsoft Research is. Microsoft Research is a part of Microsoft, but just like other research institutions at large companies, they have a great deal of autonomy. Its not like Microsoft's main teams are working on this project.
Why, it would look like . . . UNIX.
Ironic. UNIX was designed primarily for simplicity. It was a reaction to a system (MULTICS) that was highly secure, but very complex. UNIX is not a particularly secure OS design. It's simply one that doesn't do anything dumb with regards to security (as DOS/Win9x did, and as later versions of NT did).
A joke, at least at first glance.
And what exactly is funny about it? Sure, Microsoft's main programming teams have a laughable record of security, but Singularity isn't developed by rank and file programmers, its developed by a team of researchers.
If you read the press releases Microsoft is trying to bill Singularity as a research project, whose components might be used in future products.
Which is precisely what it is.
compete with Linux/Unix in the server and embedded space and craft a replacement for its aging Windows OS.
First, Windows is hardly aging. NT is fairly young as OSs go. Second, Windows cannot be replaced, not by something like Singularity. If the author actually understood the design of Singularity, he'd realize that.
Remember, Vista was first started around 2001, so assuming it ships in 2007 (which is not unrealistic)
Earlier than that. Vista traces many of its technologies back to features that were planned, in the mid 1990s, to ship with NT 5.0.
They should be looking at this as the foundation for a better, more stable, more easily maintainable base from which to build their next generation of operating systems.
They should also look at it for what it is --- an OS too different from existing systems to be a commercial success.
Because of the limited use of C# and Sing# as compared to C++ and Java Microsoft might ultimately be making a mistake in choosing Sing# as their foundation language.
Well, he's right in that they should be using Lisp (but then Singularity would be Genera :-), but of the popular choices, an extended C# is the best one. C++ just plain wouldn't work (Singularity depends on a dynamically safe language to enforce its protection system), and C# wouldn't work because it has no provisions for unsafe operations (for use in kernel code and drivers).
The rest of the implications are merely tautological ("Singularity might or might not succeed" --- duh!).
Leaving the article aside, I should point out that Singularity is a very interesting system from a theoretical point of view. It's protection model is interesting (though not unique), in that it uses the safety guarantees of the language (in this case, C# running on the .NET CLR), to protect processes from each other. This allows for a very fine grained security and sharing model. Processes can basically only access objects to which they have a reference. Since references cannot be constructed (no pointer arithmetic), they must be passed to the process from another process that owns the object.
This mechanism is at once very flexible, and very lightweight. It also obviates significant machinery in the OS and hardware. The OS's VM goes back to simply handling swap --- instead of handling swap, shared memory, and file I/O as it does in modern UNIX's. At the hardware level, page sizes no longer need to be small (as pages are no longer the fundamental unit of sharing), so the OS can use large pages (2MB instead of 4k, for example), which are adequate for swapping purposes, but not necessarily sharing purposes. This decreases VM page-management overhead, as well as improves TLB efficiency by allowing the same number of TLB entries to cover a larger working set.
Member since:
---
Member since:
---
Member since:
2005-07-11
A new MS OS without dependence on older versions. Still, will this you could make it run older software.
After all, GNU/Linux, BSDs and other *nix systems running on x86 CPUs can run Wine ( http://winehq.com/ ) and hence run a great deal of Windows applications. So certainly, MS could take this approach, and as they have details about the inner workings of MS Windows, they should be able to implement it quite well.
Old sand-boxed applications, security in mind, ..., It would be like running MS Windows Applications in Wine sandboxes on GNU/Linux.
This indeed would be interesting.
Member since:
2005-07-13
Member since:2005-07-09
Member since:---
Member since:
2005-07-08
This is truly a very bad article. Rayiner pointed out a bunch of specific flaws, but I think to biggest flaw is that the article draws conclusions that seem (to me) to be exactly the reverse of what they really are.
The article suggests that Singularity's technical details are pretty much a "joke" or something more-or-less UNIX-like. Actually, the ideas are quite innovative particularly the use of SIPs as the process model. This isn't a joke, and it definitely isn't even remotely UNIX-like.
The article also suggests that Singularity represents MS's plans for an NT replacement 10-years down the line. Or a possible exit strategy at the very least. The author clearly doesn't understand how large corporate software research labs work. Most of the projects never see the light of day, are used only internally, or merely contribute ideas/results to the development of real products. Singularity might enable MS to develop SIP technology or advanced IPC channels in real products (i.e. an NT-based OS). But it won't make a clean jump from MS Research to the MS Platforms divisions.
When it comes to radical new platform designs, I have two comments: I think we are long overdue for new concepts in computing platforms, and I don't think it's going to happen in the next 5-10 years. As much as I would consider it a failure of the computer science minds of the last 35 years, I think a predominantly UNIX-derivative-based computing market is the best we can hope for over the next 10-years. Blame IBM, AT&T, and Sun (not necessarily in that order).
Member since:2005-07-06
I say blame the DOJ for breaking up Bell, and thus absolutely and totally destroying the most important and inventive research facility during the 20th century (if not of all recorded history).
Now we have Lucent Technologies... A company that used to be Bell Labs, the research arm of Bell, trying to stay afloat because they have to come up with products to sell (like a damn answering machine).
So Lucent, formerly Bell Labs, is trying to make money by selling answering machines (and other equipment) instead of researching new technologies (like the transistor).
How different would our world be if Bell had not been broken up? Today we still have a monopoly on phone services (Verizon), and phone service which is STILL of lesser quality than what Bell used to offer. We have Verizon with it's insanely high prices (rivaled only by Time-Warner's digital phone service).
Sorry, for some reason the whole failure in computere science line brought out some emotion!
Member since:2005-12-13
Member since:2005-07-09
Member since:2005-07-06
It's not really a failure of computer science so much as it is a failure of the market to be able to absorb new technologies in the presence of existing "good enough" technologies. Of course, that's more a reality of life than a particular failure.
Computer science has come up with lot's of interesting ideas. SAOS's (single address space operating systems) protected by language safety is a very old one (dating back decades). Many of the things you're seeing in the commercial world now are state of the art technologies in the academic world --- circa 1985. The whole vector graphics UI thing, for example. In the realm of computer languages, Java and C# haven't even caught up to the sophistication (and performance!) that Smalltalk and Lisp systems reached between 1985 and 1995. In the area of user interfaces, there is a lot of work on constraint-based layout (ie. a much more powerful generalization of the automated layout models of GTK+ or Java) that are still more or less completely untapped. The current state of the art in computer science, stuff like proving the safety of parts of computer programs, won't see the light in the commercial world for a decade or two, at least.
The ideas are there, it's just that the path from academia/research to commercial production is just much longer in the computer field than it is in many others.
Member since:2005-07-06
The article also suggests that Singularity represents MS's plans for an NT replacement 10-years down the line. Or a possible exit strategy at the very least. The author clearly doesn't understand how large corporate software research labs work. Most of the projects never see the light of day, are used only internally, or merely contribute ideas/results to the development of real products. Singularity might enable MS to develop SIP technology or advanced IPC channels in real products (i.e. an NT-based OS). But it won't make a clean jump from MS Research to the MS Platforms divisions.
True, but at the same time, things can take a while; look at how things are developing now, they're pushing WinFX as a replacement for Win32, and you'll eventually start to see an eventual, but smooth, migration to completely menaged applications - when this is done, this is where Singularity and its research may come in order.
True, we may not see a pure Singularity operating system just as Windows NT was never going to remain the pure engineering feet that put correctness before buzz words - sooner of later, it will get corrupted by some 'wet behind the ears' programmers - whether it survives their buggery, is to be seen.
When it comes to radical new platform designs, I have two comments: I think we are long overdue for new concepts in computing platforms, and I don't think it's going to happen in the next 5-10 years. As much as I would consider it a failure of the computer science minds of the last 35 years, I think a predominantly UNIX-derivative-based computing market is the best we can hope for over the next 10-years. Blame IBM, AT&T, and Sun (not necessarily in that order).
I wouldn't blame them, its hard to blame a company who is simply ignorant - SUN is in the best position, but what they need is someone who understands the desktop, to bring Solaris, desktop wise, kicking and screaming to the mainstream - making it an accessible option for those wanting a Windows replacement, and want the whole kit and kaboodle supplied to them - the workstation, operating system and support infrastructure.
Member since:
2005-07-06
If you look the sixth chapter of the document you learn that the current Windows beats Linux in performance and memory usage. Did you know that "on the Unix systems, the programs are statically linked to bring in their libraries" as the document tells?
For me the Singularity operating system seems to be currently only way to spread Microsoft marketing propaganda.
Member since:2005-07-06
Congratulations on missing the point of the document entirely. It did not say that "on Unix, programs are statically linked", rather that, in their test, programs were statically linked. But that wasn't the point of the test. The point of the test was to show that Singularity's overhead for accessing system services is an order of magnitude lower than Linux's or Windows, which should be the case, because both have to deal with very expensive transitions between hardware protection domains.
Member since:
2005-07-01
I'd be curious to see what would happen if this project was later used as the basis for a windows replacement in the commercial environment.
Its about time the IT sector was shaken to its foundations with some truly innovative software products. Its ironic that Microsoft, of all companies could be the ones to bring it (a second time, one might say).
Member since:
---
Reading this article my QI dropped 10 points. The comments are even worse:
The only difference is that Apple is using UNIX, a base already created by someone else.
And MS want to make a ‘UNIX’ of their own.
IMO it’s a waste of time, they should buy a distro of LINUX and create a modular OS from there, pretty much like Apple.
Why do people think they are qualified to talk about operating system design?
Member since:---
Your QI dropped 10 points?
Granted the article could be better. Of course my hope for vista was a complete rewrite trashing all backward compatiblity, and then use something like wine, or virtualPC running a specific version of windows.
That hasn't happened Of course MSFT screwed it up. Singularity if it ever get's going will be trashed by MSFT marketing as well.
Member since:2005-07-06
Well, tell me, what exactly is WINE?
It is a project that has implemented a win32 system on *nix. It's not run in a separate environment (Then it would be an emulater).
This is no different than how Windows incorporates APIs. Win32 is just a subsystem (Same goes for the 16bit subsystem). So to have Vista run Win32 apps in a WINE environment makes it exactly the same. This is the most seamless way to do it.
Running the old apps through VirtualPC is the only way to protect the OS from these older, crappy win32 apps. But then you have the whole mess that Apple had with running OS9 over OSX. It is just such a hassle. And so much for all those games that are out, because they won't do well through VirtualPC.
It is really just a very difficult thing to do with such a huge marketshare. I think the best they can do is just to get all developers to start developing towards the new APIs, perhaps even get them to rewrite older and important apps to it, and then just phase out Win32 (also note that there is still a win16 subsystem in Windows as well, so this could take a VERY long).
Member since:2005-07-06
Member since:
---
Member since:
2005-07-11
I agree this is a terrible article, but you have to remember that this is targeted at Apple type readers, and that this is the type of information that they are reading and absorbing and believing in on a daily basis.
The purpose of Singularity is to provide practical information about the performance and security implications, advantages and disadvantages, and new possibilities as the managed framework (.net) inevitably travels deeper into the OS core framework.
Those interested might like to take a look at projects like http://www.jnode.org/ and http://www.jdistro.com/ - these are OS projects targeting java.
Member since:
---
In particular language type safety as an alternative to operating system level process protection seems to be a feature of the Oberon (cooperative multitasking system.) Systems where processes resemble threads running in the same memory process and all objects can quickly send messages to each other have been around for years (like Smalltalk.)
Member since:
2005-07-06
yes thats the word im going to use about singularity.
the idea that its not the individual app that manages its own memory, but the very framework of the os that does so should help in security as the single most common security problem these days is a unmanaged buffer on some server.
yes, there will be a performance hit. but face it, this isnt the early 80's. today you cluster together desktop cpu's to get the horsepower you want, not buy a custom chip inside a custom big iron machine from the likes of ibm or cray (alltho you can still do so when a cluster isnt the right way to attack the problem).
and thanks for the inferno link, ill look into that 
edit:
hmm, i see the term namespace pop up over and over. and i keep coming back to plan9. i realy do wonder why plan9 never took of as it seems that it allready contains all the stuff that singularity and others are supposed to contain.
im guessing its all to do with licence. is plan9 under the same mit/bsd licence that the original unix code was under? if not then things start to make sense and yet again point towards what the combo of licences and patents can lead to.
basicly mom and pop corp is very afraid of letting their techs look at a diffrent os's codes and features in case they get slapped around with a lawsuit. what would be the basis of said suit? that a feature of the latest os from mom and pop corp looks suspiciusly similar to another os.
because one implementation of the wheel is under patent or licence one keeps recreating it, maybe badly, maybe not, but still its recreated rather then buildt upon...
Edited 2005-12-13 13:18
Member since:
2005-07-06
Member since:
---
This is just my (admitedly largely uninformed) opinion, but I think that probably the thing for Microsoft to do WRT Windows is to keep it based on the NT kernel, but to re-write everything on top of it with completely managed .NET code, and run legacy applications in a virtualized environment.
While going this route would not give them the kernel level advantages of Singularity (lessening the reliance on the MMU for various low-level tasks), it would make make most of the OS and all of the applications much safer, and from what I've heard, easier to properly write in the first place.
Of course, this isn't likely going to happen, as Microsoft as well as all third party developers have made a huge investment in Win32. It's a shame though, when you think about the advantages that .NET offers. No dealing with pointers, directly dealing with menory (far fewer memory leaks etc.) buffer overflows become virtually non-existant and so on.
Hell, I'd be happy enough (for the time being) if Microsoft would even just re-write the Win32 functionality that .NET currently ony wraps around, so that all fully .NET apps did not ever need to touch Win32. If only Microsoft had spent the last few years doing this in Longhorn/Vista instead of whatever it was they were in fact doing, the future of Windows might actually be brighter than just a flashy new look and some tacked on technologies.
(yeah, I know that quite a bit of their new features are implemented in .NET but you know what they say about building castles on swamps... )
Member since:
2005-11-11
"Old sand-boxed applications, security in mind, ..., It would be like running MS Windows Applications in Wine sandboxes on GNU/Linux."
I get your point, no reason MS couldn't have a sandboxed emulator for old apps. It is important to remember though that apps running in Wine are not running in any sort of sandbox.
The only thing preventing bad stuff from happening is that apps run in Wine tend to be unaware of the structure of the Linux around them, but if one was, it could run amok and erase your personal configuration files or whatever.
"If you run untrusted code under your account it can do anything that you are allowed to. This is exactly equivalent to running an untrusted Linux app. From a security standpoint there is absolutely no difference between a Windows binary running under Wine and a Linux binary running natively. You can use the DOS drive configuration to limit the potential problems a bug in a Windows app can cause; but it is impossible to protect against malicious code except by not running it. Wine is not, and cannot be, a sandbox for running untrusted code."
http://www.winehq.org/?issue=142
http://www.winehq.org/hypermail/wine-devel/2002/10/0803.html
Member since:
2005-07-06
Member since:2005-07-06
Member since:
---
- microkernel
- uses isolated, protected processes
- it is their attempt to build a "Trusted Computing" platform
- it is written in Sing#, and extension of C# which is compiled into Microsoft Intermediate Language, which is what their Common Language Runtime target language is (.Net!). There is then a compiler which compiles the MSIL into x86 instructions - so there could be one for any architecture, really (MS on Mac!)
- library version - they claim it will create a strong path for system evolution and backward compatibility (MS loves this!) - this includes versioning for the ABI (i.e, application binary interface) to the kernel
- mention at the bottom of page 30 that performance is good, yet they have no way to demonstrate the actual goal of dependability - then go on to show performance...I don't get what the purpose for this was. It blows the abstract out of the water
- mentions other state of the art OSes including Inferno which is really Plan 9 -> or a better unix from the guys who buit the original one
Over all, I have no opinion except for the decision to use .Net as the basis for the OS. I will be following this closely.
Member since:
---
This is just an excercise in "take your favourite non system programming language, java, lisp, perl, rebol or whatever and try to write an OS using it"
The end effect generally performs rather worse than the same system written in Assembler and C and you end up with snarfed up ideas like run everything as an object soup in a common memory space...
Member since:---
"The end effect generally performs rather worse than the same system written in Assembler and C and you end up with snarfed up ideas like run everything as an object soup in a common memory space..."
is it not supossed that common memory space does help to performance? is it not the ability to touch the memory of other processes without context switching the excuse to make thing like puting graphics server in privileged ring levels?
anyway, yes the thing is old idea. The problem i see is many projects like this is thay tend to choose a language and with this they put many people out. At least here ther is a minimum common assembly lang and many languages compiling to it, so they can "please" more developers. The same can be said for jnode.
Member since:2005-07-06
Said like a C programmer.
Life is a lot more complicated than you think. Performance is not a matter of how something does on a microbenchmark (which Lisp, Java, etc, will generally always lose) but how the system performs as a whole. C is great for microbenchmarks, but its lack of safety can result in issues that have adverse affects on system performance.
Consider, for example, the perrenial GUI system decision: graphics in the kernel or graphics in userspace? This topic has been the subject of many debates, but in a safe system, it becomes a moot point. In such a system, the graphics server has the performance of a kernel-level system, and the safety of a userspace one.
Consider X. Much of the apparent slowness in X's performance comes not from raw drawing speed (which is more than fast enough for most things), but synchronization. Synchronization in X platforms isn't implemented very well. Why? Because there are three seperate processes involved --- the window manager, the X server, and the client. Synchronizing these well can become quite complicated and result in a lot of overhead. Now, the primal C programmer will blame this situation on X as a system, but that's because he fails to understand the issues involved. X is made up of little programs largely for flexibility and safety. C's support for in-process dynamic code loading and security ranges from primitive to non-existent. X is broken up to address these problems. In a system like Singularity, it wouldn't matter. X could be just a userspace library, and you'd retain the full flexibility and security of the seperate-process model. Performance would improve, not just because of reduced IPC costs, but because of massively simpler synchronization.
Existing systems are faced with a great problem. System calls and IPC are very expensive, and the costs are only getting worse as processors get more complex. Some systems ignore these costs (Mach), and were dismal failures for it. Others go to great lengths to hide these costs (modern microkernels, most monolithic systems). There is elaborate user-space buffering, userspace locking mechanisms (futexes on Linux), userspace libraries with kernel space drivers (DRI/DRM), etc. In a system like Singularity, all of these things go away. Who cares if there is a constant 20% speed hit from a safe language? Much better a constant 20% hit than a non-linear hit caused by limitations in locking, synchronization, etc.
Member since:
---
I find it facinating that we accuse MS (oops, I mean, MS) of not innovating, and then when they do with a new research OS like this, we kick them for that too.
Singularity has some great new concepts in there, and it's more than Sun or IBM, or the open source community at large, has given us.
Timbert
Member since:
---
Member since:---
Member since:
---
Uhm. This article is poor. Why did you link to it here? Clearly, the author has little understanding of what's behind Singularity. It's a research kernel. How could a closed, micro-kernel, SIP-based, Channel-based system replace Windows? I mean, you'd think Microsoft would want Windows apps to still be able to run. What a joke. Next();
Member since:2005-11-15
Member since:
---
I don't think the author of this article knows anything about Singularity. He doesn't substantiate any of his statements and thinks it is *nix because of the screen shot; for anyone who has read or watched Singularity videos it's clearly much more than a *nix clone. As for the future of it he could be right or wrong, but again, no substance to prove he's right.
Member since:
2005-06-29
Member since:
---
You believe that because you never studied computer architecture or system, otherwise you would just say that it is unbelivable that an OS need 600 MO, a Kernel 190 MO, an WM 200MO
All this stuff should require only some KO, sould be fast and save.
MS is following this way with Singularity OS, that s a good point for them.
anyway vista will be a big shit, just like linus is one and xp too.
Member since:
2005-07-12
To be honest, I was wondering when Microsoft would dust off XENIX to try and squash the "upstart" Linux.
Something people forget at times, Microsoft actually used to make a UNIX varient... that a hefty portion of was used as the launching point for the creation of NT 3.5
Edited 2005-12-14 12:19
