On Tuesday 12 February 2008 17:08, Matthew Toseland wrote:
> AFAICS this is the way forward. Since Java attempts to use POSIX priorities
> and they don't do anything on Linux, and since thread priorities work on
> Linux (and presumably on Solaris), it shouldn't result in any drastic bugs.
What exactly would we need to do to get this working? AFAICS we will need to
ship a binary library for Linux for each processor type we support? Can you
give me a makefile/shell script to build the below? It needs to be easy to
build such libraries for a new platform, although we will keep them in binary
form in freenet-ext.jar (built on trustworthy dev's machines).
>
> On Sunday 10 February 2008 05:21, Jack O'Lantern wrote:
> > Matthew Toseland <toad at amphibian.dyndns.org> wrote:
> > > Beyond that, the only realistic option appears to be an external daemon
> > > running at a lower nice level to do CPU-intensive jobs and FEC encoding
> and
> > > decoding in particular.
> >
> > Thread.yield() comes to mind. But this is ugly.
> >
> > Wouldn't it be sufficient if the java threads set their own nice level?
> Instead
> > of shoving it all off to a daemon? This is quite trivial, and it might be
> > worthwhile to ask the java people to have Thread.setPriority do just that
as
> > long as Linux NPTL does not support SCHED_OTHER prioritizing. We might
even
> do
> > it ourselves. Here is something quick and dirty:
> >
> > public class NativeThread extends Thread {
> >
> > private static final int PRIO_RANGE = MAX_PRIORITY - MIN_PRIORITY;
> > private static final int NATIVE_PRIO_RANGE = 19;
> >
> > /**
> > * set linux priority.
> > *
> > * @return true if successful, false otherwise.
> > */
> > private native boolean setLinuxPriority(int prio);
> >
> > /**
> > * rescale java priority to linux priority and
> > * set linux priority.
> > * On failure, fall back to standard setPriority()
> > */
> > public void setNativePriority(int prio) {
> > int nPrio = NATIVE_PRIO_RANGE - (NATIVE_PRIO_RANGE *
(prio -
> > MIN
> > _PRIORITY)) / PRIO_RANGE;
> > if (setLinuxPriority(nPrio))
> > return;
> > else
> > System.err.println("Setting Linux priority
> failed.");
> > setPriority(prio);
> > }
> >
> > static {
> > System.loadLibrary("NativeThread");
> > }
> > }
> >
> > The native setLinuxPriority() would look like this:
> >
> > #include<sys/resource.h>
> > #include<sys/time.h>
> > #include<stdio.h>
> >
> > #include"NativeThread.h"
> >
> > JNIEXPORT jboolean JNICALL Java_NativeThread_setLinuxPriority
> > (JNIEnv * env, jobject jobj, jint prio) {
> > if (setpriority(PRIO_PROCESS, 0, prio) == 0)
> > return JNI_TRUE;
> > return JNI_FALSE;
> > }
> >
> > I have tested both NativeThread and Thread.yield():
> >
> > public class NativeThreadTest {
> >
> > static class ThreadA extends NativeThread {
> >
> > public void run() {
> > setNativePriority(MAX_PRIORITY - 1);
> > while (true)
> > System.out.print("A");
> > }
> >
> > }
> >
> > static class ThreadB extends NativeThread {
> >
> > public void run() {
> > setNativePriority(MAX_PRIORITY);
> > while (true)
> > System.out.print("B");
> > }
> >
> > }
> >
> > public static void main(String[] args) {
> > Thread ta = new ThreadA();
> > Thread tb = new ThreadB();
> > Thread tm = Thread.currentThread();
> >
> > ta.start();
> > tb.start();
> >
> > while (true) {
> > System.out.print("M");
> > tm.yield();
> > }
> > }
> >
> > }
> >
> > So the main thread (call it "M") creates to new threads "A" and "B" which
> set
> > their priority to MAX-1 and MAX, respectively. All threads then print
> > characters continuously, with M constantly yielding. Now B gets indeed the
> most
> > CPU time, and the yielding M gets almost nothing. I performed this test
with
> > with other priorities than MAX-1 for A and found the outcome to be very
> > sensitive to these changes. In particular, when A's priority was set to
MIN,
> it
> > did not print a single character in several minutes.
> >
> > Of course, this is only a rather crude test, so you might want to
reconcile
> the
> > outcome with data from your own testbed.
> >
> > Jack
> >
> >
> >
>
____________________________________________________________________________________
> > Be a better friend, newshound, and
> > know-it-all with Yahoo! Mobile. Try it now.
> http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
> >
> >
> > _______________________________________________
> > Devl mailing list
> > Devl at freenetproject.org
> > http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
> >
> >
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL:
<https://emu.freenetproject.org/pipermail/devl/attachments/20080212/8afa71a2/attachment.pgp>
