Linux Gets O(1) SMP Patch As Late Christmas Gift

bodin writes: "Now that new-year's parties are over things are getting boring again. For those who want to see and perhaps even try something more complex, Ingo Molnar is announcing this patch that is a pretty radical rewrite of the Linux scheduler. This is big stuff!"

Comments

O(1) == constant time

[adadun]

For those of you who haven't read enough computer science to know what O(1) means, here is a short explanation.

The Big O-notation is a way to describe how the (asymptotic) execution time of an algorithm depends on the inputs to the algorithm. For instance, an algorithm that loops over n values is said to have the asymptotic execution time O(n) - it is proportional to the number of times the loop is executed.

Similarly, an algorithm that runs in constant time, i.e., that takes equally long to execute for 10 values and for 1000000 values, is said to be O(1). The execution time is proportional to 1.

For the Linux scheduler, switching processes is O(p), where p is the number of currently running processes. This new scheduler switches processes in O(1) time.

This means that even though the old scheduler might be fast for low numbers of running processes, it will take longer and longer timer to switch processes when the number of active processes grow. The new scheduler will switch processes just as fast for 2 processes or for 200 processes. Even though the new scheduler might be slower in when there are few processes, it will be faster when the number of running processes increase.

Playing with patch.......

[orangesquid]

By the way, since the links /and/ the URL's (as far as I can tell) in the post are broken, this should help:

http://people.redhat.com/mingo/O(1)-scheduler/

I've created diffs between 2.5.1 and 2.5.2pre8 with the O(1) scheduler, and between 2.5.2pre8 and 2.5.2pre8 with the O(1) scheduler.

2.5.2pre8 actually patches pretty well with the original scheduler patch (drivers/char/serial.c.rej can be ignored, and you have to make a few changes to kernel/sched.c in order for it to patch correctly), but because it took me at least ten minutes of fiddling with sched.c I've decided to make a diff for 2.5.2pre8.

No guarantee that either of these works, though :) but hey, my kernel compiled just fine.

# diff -ru linux-2.5.1 linux-2.5.2pre8schedO1|grep -v '^Only in '|gzip -f >/home/web/patch-2.5.1-2.5.2pre8schedO1.patch.g z


http://os.markbach.com:8080/patch-2.5.1-2.5.2pre 8s chedO1.patch.gz
(396,961 bytes)

# diff -ru linux-orig linux |grep -v '^Only in ' |gzip -f >/home/web/patch-2.5.2pre8-2.5.2pre8schedO1.pat ch.gz


http://os.markbach.com:8080/patch-2.5.pre8-2.5.2 pr e8schedO1.patch.gz
(31,124 bytes)

Good luck to anyone who tries to use these :)

And no, I didn't patch in the kdev_t stuff from people/aeb on kernel.org because there's lots of kdev_t stuff in the Changelog for pre7 and pre8, so I decided to assume (yes, I know, assuming makes an ass out of u and me) I didn't need it... of course, when the system crashes after five seconds, maybe I'll change my mind ;)

And if, for some odd reason, you can't connect on port 8080, just connect on port 80 and let's hope you're not blocked by @home's or my firewall. :)

Damn, I'm using too many smileys :-\

Re:What I would like to know.

[Anonymous Coward]

Apache 1.x does in fact pool processes, as does the 2.0 prefork MPM. All processes save one block on a semaphore, and only the one gets into listen (2).

The real reason threads can be faster than processes if all other things are equal is context switch time. Switching between threads that share all their page tables is just a matter of restoring all the registers. When switching between processes, you must flush the TLB also.

Re:What I would like to know.

[pthisis]

The real reason threads can be faster than processes if all other things are equal is context switch time. Switching between threads that share all their page tables is just a matter of restoring all the registers. When switching between processes, you must flush the TLB also.

But when threads modify memory on SMP, they have to ensure cache coherency between CPUs. Processes can don't have this restriction, which is why processes can often be faster than threads. It all depends on your usage patterns and OS. On e.g. Solaris or Win32, threads are far faster at context switching than processes (mainly because processes are so heavyweight on those systems; Win32 also doesn't offer an API to efficiently implement multiprocess programs). On identical hardware, Linux's processes switches are faster than Solaris or Win32 thread switches and Linux thread switches aren't that much faster than process switches (beacause process switching is implemented in an efficient manner).

So on Linux, the fact that memory isn't shared between processes (hence CPU coherency isn't the same issue) often makes processes a substantial performance _win_.

This is aside from the fact that threads are extremely difficult to program properly--there's a reason that OSes spent all that time implementing protected memory (remember all the hoopla about that?) and threads throw that concept out the window, plus tend to lure the programmer into architectures that need nasty locking and synchronization.

That said, for some applications threads are the right answer. But using multiple processes, possibly with shared memory for a limited set of data, is generally a more maintainable solution that's easier to architect and implement. It certainly makes it explicit which data structures are meant to be shared, which a multithreaded solution does not.

Sumner

Re:Scheduling details: switching process != O(p)

[sorbits]

Picking a process to run is O(p) as long as no real-time processes are runnable (which is the vast majority of the time)

ehm... most of the time there are no processes to run, they all wait for a message/interrupt. When one arrives, they are moved to the list (priority queue) of running/ready processes.

The schedular always just pick the next in this list of running/ready processes, which is O(1). At no time does it need to look at the iddle processes, to pick one to run, as they don't need to be run... this is not Microsoft's cooperative multitasking...