The Microsoft Singularity

jose parinas writes ""Microsoft Research has published the first details of a wholly new operating system under development called Singularity, designed new from the ground up, built on a new language and designed with emphasis on dependability instead of performance.""

Comments

I hacked on this...

[megabeck42]

I saw and worked on this a bit while interning at Microsoft. Although what I say is my own and doesn't reflect Microsoft in any way, it's important to remember that this is a research operating system, so its not challenging or replacing Windows. They have some very good, solid ideas. I hope that, someday, it will be released.

Re:another longhorn?

[Overly Critical Guy]

This is just a research OS written in C#.

Microsoft Research is always making things Microsoft never uses. Remember all the 3D navigator stuff they were crowing about years ago?

I think Microsoft Research is a place to keep eggheads working and happy so they don't go working somewhere else.

Re:another longhorn?

[Anonymous Coward]

Come to think of it, has an written from scratch OS worked?

CP/M. Some device specific OS's like the C64. The Apple Lisa (which begat the MacOS 1-9). I'm not even sure I'd count Unix, since there's so many flavors all based on something from before. Even Linux was heavily dependant on GNU, which pre-existed it.

Look at the failures. BeOS. Rhapsody. Plan 9, etc.

Re:My guess:

[zootm]

If you read the paper, the idea is that, yes, it'll be slower, but the reliability will be built in from the beginning, rather than other systems which take something fast and bolt reliability on. They make a good point that they will be able to use optimising compilers for CLR languages in this context, too.

apples and oranges

[idlake]

The two are very different beasts.

EROS uses C and relies on memory management hardware for isolation. EROS also can't analyze or verify code it loads.

Singularity uses C# and does not use memory management hardware for protection; it guarantees isolation via runtime checks, and it can perform extensive code analysis on load.

I don't know whether Singularity is going to make it, but I have used and developed on systems like it (the idea isn't new), and it is a lot nicer than either UNIX kernels or EROS-like kernels.

Re:IMPORTANT

[sstidman]

I'll admit that I don't know much about kernels, but I'll try to summarize. The chart compares performance between the new OS Singularity, FreeBSD 5.3, Redhat Fedora Core 4 (kernel version 2.6.11-1.1369_FC4), and Windows XP (SP2). The goal of the chart, stated in the paragraphs above, is to show that the new Singularity architecture does not suffer any performance hits in order to make a more secure system.

      The table shows the CPU cost of six different types of operations: "Read Cycle Counter", "ABI Call", "Thread yield", "2 thread wait-set ping pong", "2 message ping pong" and "Create and start process". For the first one, Windows seems to kick the butt of all others handily with Singularity being the worst of the bunch. For "ABI Call", each OS used different system calls that "operate on a readily available data structure in the respective kernels." The system calls seem to be completely different so I don't know if this test is valid, but the results show Singularity an order of magnitude more efficient than the others, with Linux beating Windows by a considerable margin and Windows beating FreeBSD by an equally considerable margin.

      For the "thread yield" tests, FreeBSD & Linux are equal, Windows beats them by a reasonable percentage and Singularity is more than twice as fast as the Unixes. For the "2 wait-set ping pong", which measures "the cost of switching between two threads in the same process through a synchronization object", the chart shows that Singularity is somewhat more efficient than Windows and Windows is more than twice as fast as the Unixes. For the "2 message ping pong", which shows the cost of sending a 1-byte message back and forth between two processes, Singularity is 4 times more efficient than Linux, which is somewhat better than Windows, which kicks FreeBSDs butt decisively.

      Lastly, for "Create and start process", Singularity is twice as fast as Linux, which is about 50% faster than FreeBSD. Windows comes out 7 times slower than Linux on this test. I don't know how much that matters in the real world since creating and starting a process is not something that is done hundreds of times a second.

      All that said, it should probably be pointed out that there are many ways to measure an OS. The M$ guys may have simply picked the ones that support their "see we don't suck" position. And given that Singularity is not a complete OS, I would expect that more overhead will be added later that will bring down these numbers. I guess we'll see.