scheduler.cc

Go to the documentation of this file.

/* -*-  Mode:C++; c-basic-offset:8; tab-width:8; indent-tabs-mode:t -*- */
/*
 * Copyright (c) 1994 Regents of the University of California.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *  This product includes software developed by the Computer Systems
 *  Engineering Group at Lawrence Berkeley Laboratory.
 * 4. Neither the name of the University nor of the Laboratory may be used
 *    to endorse or promote products derived from this software without
 *    specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 * @(#) $Header: /nfs/jade/vint/CVSROOT/ns-2/common/scheduler.cc,v 1.73 2005/08/22 05:08:32 tomh Exp $
 */

#ifndef lint
static const char rcsid[] =
    "@(#) $Header: /nfs/jade/vint/CVSROOT/ns-2/common/scheduler.cc,v 1.73 2005/08/22 05:08:32 tomh Exp $ (LBL)";
#endif

#include <stdlib.h>
#include <limits.h>
#include <math.h>

#include "config.h"
#include "scheduler.h"
#include "packet.h"


#ifdef MEMDEBUG_SIMULATIONS
#include "mem-trace.h"
#endif

Scheduler* Scheduler::instance_;
scheduler_uid_t Scheduler::uid_ = 1;

// class AtEvent : public Event {
// public:
//  char* proc_;
// };

Scheduler::Scheduler() : clock_(SCHED_START), halted_(0)
{
}

Scheduler::~Scheduler(){
    instance_ = NULL ;
}

/*
 * Schedule an event delay time units into the future.
 * The event will be dispatched to the specified handler.
 * We use a relative time to avoid the problem of scheduling
 * something in the past.
 *
 * Scheduler::schedule does a fair amount of error checking
 * because debugging problems when events are triggered
 * is much harder (because we've lost all context about who did
 * the scheduling).
 */
void 
Scheduler::schedule(Handler* h, Event* e, double delay)
{
    // handler should ALWAYS be set... if it's not, it's a bug in the caller
    if (!h) {
        fprintf(stderr,
            "Scheduler: attempt to schedule an event with a NULL handler."
            "  Don't DO that.\n");
        abort();
    };
    
    if (e->uid_ > 0) {
        printf("Scheduler: Event UID not valid!\n\n");
        abort();
    }
    
    if (delay < 0) {
        // You probably don't want to do this
        // (it probably represents a bug in your simulation).
        fprintf(stderr, 
            "warning: ns Scheduler::schedule: scheduling event\n\t"
            "with negative delay (%f) at time %f.\n", delay, clock_);
    }

    if (uid_ < 0) {
        fprintf(stderr, "Scheduler: UID space exhausted!\n");
        abort();
    }
    e->uid_ = uid_++;
    e->handler_ = h;
    double t = clock_ + delay;

    e->time_ = t;
    insert(e);
}

void
Scheduler::run()
{
    instance_ = this;
    Event *p;
    /*
     * The order is significant: if halted_ is checked later,
     * event p could be lost when the simulator resumes.
     * Patch by Thomas Kaemer <Thomas.Kaemer@eas.iis.fhg.de>.
     */
    while (!halted_ && (p = deque())) {
        dispatch(p, p->time_);
    }
}

/*
 * dispatch a single simulator event by setting the system
 * virtul clock to the event's timestamp and calling its handler.
 * Note that the event may have side effects of placing other items
 * in the scheduling queue
 */

void
Scheduler::dispatch(Event* p, double t)
{
    if (t < clock_) {
        fprintf(stderr, "ns: scheduler going backwards in time from %f to %f.\n", clock_, t);
        abort();
    }

    clock_ = t;
    p->uid_ = -p->uid_; // being dispatched
    p->handler_->handle(p); // dispatch
}

void
Scheduler::dispatch(Event* p)
{
    dispatch(p, p->time_);
}

class AtEvent : public Event {
public:
    AtEvent() : proc_(0) {
    }
    ~AtEvent() {
        if (proc_) delete [] proc_;
    }
    char* proc_;
};

class AtHandler : public Handler {
public:
    void handle(Event* event);
} at_handler;

void 
AtHandler::handle(Event* e)
{
    AtEvent* at = (AtEvent*)e;
    Tcl::instance().eval(at->proc_);
    delete at;
}

void
Scheduler::reset()
{
    clock_ = SCHED_START;
}

int 
Scheduler::command(int argc, const char*const* argv)
{
    Tcl& tcl = Tcl::instance();
    if (instance_ == 0)
        instance_ = this;
    if (argc == 2) {
        if (strcmp(argv[1], "run") == 0) {
            /* set global to 0 before calling object reset methods */
            reset();    // sets clock to zero
            run();
            return (TCL_OK);
        } else if (strcmp(argv[1], "now") == 0) {
            sprintf(tcl.buffer(), "%.17g", clock());
            tcl.result(tcl.buffer());
            return (TCL_OK);
        } else if (strcmp(argv[1], "resume") == 0) {
            halted_ = 0;
            run();
            return (TCL_OK);
        } else if (strcmp(argv[1], "halt") == 0) {
            halted_ = 1;
            return (TCL_OK);

        } else if (strcmp(argv[1], "clearMemTrace") == 0) {
#ifdef MEMDEBUG_SIMULATIONS
            extern MemTrace *globalMemTrace;
            if (globalMemTrace)
                globalMemTrace->diff("Sim.");
#endif
            return (TCL_OK);
        } else if (strcmp(argv[1], "is-running") == 0) {
            sprintf(tcl.buffer(), "%d", !halted_);
            return (TCL_OK);
        } else if (strcmp(argv[1], "dumpq") == 0) {
            if (!halted_) {
                fprintf(stderr, "Scheduler: dumpq only allowed while halted\n");
                tcl.result("0");
                return (TCL_ERROR);
            }
            dumpq();
            return (TCL_OK);
        }
    } else if (argc == 3) {
        if (strcmp(argv[1], "at") == 0 ||
            strcmp(argv[1], "cancel") == 0) {
            Event* p = lookup(STRTOUID(argv[2]));
            if (p != 0) {
                /*XXX make sure it really is an atevent*/
                cancel(p);
                AtEvent* ae = (AtEvent*)p;
                delete ae;
            }
        } else if (strcmp(argv[1], "at-now") == 0) {
            const char* proc = argv[2];

            // "at [$ns now]" may not work because of tcl's 
            // string number resolution
            AtEvent* e = new AtEvent;
            int n = strlen(proc);
            e->proc_ = new char[n + 1];
            strcpy(e->proc_, proc);
            schedule(&at_handler, e, 0);
            sprintf(tcl.buffer(), UID_PRINTF_FORMAT, e->uid_);
            tcl.result(tcl.buffer());
        }
        return (TCL_OK);
    } else if (argc == 4) {
        if (strcmp(argv[1], "at") == 0) {
            /* t < 0 means relative time: delay = -t */
            double delay, t = atof(argv[2]);
            const char* proc = argv[3];

            AtEvent* e = new AtEvent;
            int n = strlen(proc);
            e->proc_ = new char[n + 1];
            strcpy(e->proc_, proc);
            delay = (t < 0) ? -t : t - clock();
            if (delay < 0) {
                tcl.result("can't schedule command in past");
                return (TCL_ERROR);
            }
            schedule(&at_handler, e, delay);
            sprintf(tcl.buffer(), UID_PRINTF_FORMAT, e->uid_);
            tcl.result(tcl.buffer());
            return (TCL_OK);
        }
    }
    return (TclObject::command(argc, argv));
}

void
Scheduler::dumpq()
{
    Event *p;

    printf("Contents of scheduler queue (events) [cur time: %f]---\n",
        clock());
    while ((p = deque()) != NULL) {
        printf("t:%f uid: ", p->time_);
        printf(UID_PRINTF_FORMAT, p->uid_);
        printf(" handler: %p\n", p->handler_);
    }
}

static class ListSchedulerClass : public TclClass {
public:
    ListSchedulerClass() : TclClass("Scheduler/List") {}
    TclObject* create(int /* argc */, const char*const* /* argv */) {
        return (new ListScheduler);
    }
} class_list_sched;

void 
ListScheduler::insert(Event* e)
{
    double t = e->time_;
    Event** p;
    for (p = &queue_; *p != 0; p = &(*p)->next_)
        if (t < (*p)->time_)
            break;
    e->next_ = *p;
    *p = e;
}

/*
 * Cancel an event.  It is an error to call this routine
 * when the event is not actually in the queue.  The caller
 * must free the event if necessary; this routine only removes
 * it from the scheduler queue.
 */
void 
ListScheduler::cancel(Event* e)
{
    Event** p;
    if (e->uid_ <= 0)   // event not in queue
        return;
    for (p = &queue_; *p != e; p = &(*p)->next_)
        if (*p == 0)
            abort();

    *p = (*p)->next_;
    e->uid_ = - e->uid_;
}

Event* 
ListScheduler::lookup(scheduler_uid_t uid)
{
    Event* e;
    for (e = queue_; e != 0; e = e->next_)
        if (e->uid_ == uid)
            break;
    return (e);
}


Event*
ListScheduler::deque()
{ 
    Event* e = queue_;
    if (e)
        queue_ = e->next_;
    return (e);
}

#include "heap.h"

Heap::Heap(int size)
        : h_s_key(0), h_size(0), h_maxsize(size), h_iter(0)
{
    h_elems = new Heap_elem[h_maxsize];
    memset(h_elems, 0, h_maxsize*sizeof(Heap_elem));
    //for (unsigned int i = 0; i < h_maxsize; i++)
    //  h_elems[i].he_elem = 0;
}

Heap::~Heap()
{
    delete [] h_elems;
}

/*
 * int  heap_member(Heap *h, void *elem):       O(n) algorithm.
 *
 *  Returns index(elem \in h->he_elems[]) + 1,
 *          if elem \in h->he_elems[],
 *      0,  otherwise.
 *  External callers should just test for zero, or non-zero.
 *  heap_delete() uses this routine to find an element in the heap.
 */
int
Heap::heap_member(void* elem)
{
    unsigned int i;
    Heap::Heap_elem* he;
    for (i = 0, he = h_elems; i < h_size; i++, he++)
        if (he->he_elem == elem)
            return ++i;
    return 0;
}

/*
 * int  heap_delete(Heap *h, void *elem):       O(n) algorithm
 *
 *  Returns 1 for success, 0 otherwise.
 *
 * find elem in h->h_elems[] using heap_member()
 *
 * To actually remove the element from the tree, first swap it to the
 * root (similar to the procedure applied when inserting a new
 * element, but no key comparisons--just get it to the root).
 *
 * Then call heap_extract_min() to remove it & fix the tree.
 *  This process is not the most efficient, but we do not
 *  particularily care about how fast heap_delete() is.
 *  Besides, heap_member() is already O(n), 
 *  and is the dominating cost.
 *
 * Actually remove the element by calling heap_extract_min().
 *  The key that is now at the root is not necessarily the
 *  minimum, but heap_extract_min() does not care--it just
 *  removes the root.
 */
int
Heap::heap_delete(void* elem)
{
    int i;
    if ((i = heap_member(elem)) == 0)
        return 0;
    for (--i; i; i = parent(i)) {
        swap(i, parent(i));
    }
    (void) heap_extract_min();
    return 1;
}

/*
 * void heap_insert(Heap *h, heap_key_t *key, void *elem)
 *
 * Insert <key, elem> into heap h.
 * Adjust heap_size if we hit the limit.
 * 
 *  i := heap_size(h)
 *  heap_size := heap_size + 1
 *  while (i > 0 and key < h[Parent(i)])
 *  do  h[i] := h[Parent(i)]
 *      i := Parent(i)
 *  h[i] := key
 */
void
Heap::heap_insert(heap_key_t key, void* elem) 
{
    unsigned int    i, par;
    if (h_maxsize == h_size) {  /* Adjust heap_size */
        unsigned int osize = h_maxsize;
        Heap::Heap_elem *he_old = h_elems;
        h_maxsize *= 2;
        h_elems = new Heap::Heap_elem[h_maxsize];
        memcpy(h_elems, he_old, osize*sizeof(Heap::Heap_elem));
        delete []he_old;
    }

    i = h_size++;
    par = parent(i);
    while ((i > 0) && 
           (KEY_LESS_THAN(key, h_s_key,
                  h_elems[par].he_key, h_elems[par].he_s_key))) {
        h_elems[i] = h_elems[par];
        i = par;
        par = parent(i);
    }
    h_elems[i].he_key  = key;
    h_elems[i].he_s_key= h_s_key++;
    h_elems[i].he_elem = elem;
    return;
};
        
/*
 * void *heap_extract_min(Heap *h)
 *
 *  Returns the smallest element in the heap, if it exists.
 *  NULL otherwise.
 *
 *  if heap_size[h] == 0
 *      return NULL
 *  min := h[0]
 *  heap_size[h] := heap_size[h] - 1   # C array indices start at 0
 *  h[0] := h[heap_size[h]]
 *  Heapify:
 *      i := 0
 *      while (i < heap_size[h])
 *      do  l = HEAP_LEFT(i)
 *          r = HEAP_RIGHT(i)
 *          if (r < heap_size[h])
 *              # right child exists =>
 *              #       left child exists
 *              then    if (h[l] < h[r])
 *                      then    try := l
 *                      else    try := r
 *              else
 *          if (l < heap_size[h])
 *                      then    try := l
 *                      else    try := i
 *          if (h[try] < h[i])
 *              then    HEAP_SWAP(h[i], h[try])
 *                  i := try
 *              else    break
 *      done
 */
void*
Heap::heap_extract_min()
{
    void*   min;                  /* return value */
    unsigned int    i;
    unsigned int    l, r, x;

    if (h_size == 0)
        return 0;
    min = h_elems[0].he_elem;
    h_elems[0] = h_elems[--h_size];
// Heapify:
    i = 0;
    while (i < h_size) {
        l = left(i);
        r = right(i);
        if (r < h_size) {
            if (KEY_LESS_THAN(h_elems[l].he_key, h_elems[l].he_s_key,
                      h_elems[r].he_key, h_elems[r].he_s_key))
                x= l;
            else
                x= r;
        } else
            x = (l < h_size ? l : i);
        if ((x != i) && 
            (KEY_LESS_THAN(h_elems[x].he_key, h_elems[x].he_s_key,
                   h_elems[i].he_key, h_elems[i].he_s_key))) {
            swap(i, x);
            i = x;
        } else {
            break;
        }
    }
    return min;
}


static class HeapSchedulerClass : public TclClass {
public:
    HeapSchedulerClass() : TclClass("Scheduler/Heap") {}
    TclObject* create(int /* argc */, const char*const* /* argv */) {
        return (new HeapScheduler);
    }
} class_heap_sched;

Event* 
HeapScheduler::lookup(scheduler_uid_t uid)
{
    Event* e;
    
    for (e = (Event*) hp_->heap_iter_init(); e;
         e = (Event*) hp_->heap_iter())
        if (e->uid_ == uid)
            break;
    return e;
}

Event*
HeapScheduler::deque()
{
    return ((Event*) hp_->heap_extract_min());
}

/*
 * Calendar queue scheduler contributed by
 * David Wetherall <djw@juniper.lcs.mit.edu>
 * March 18, 1997.
 *
 * A calendar queue basically hashes events into buckets based on
 * arrival time.
 *
 * See R.Brown. "Calendar queues: A fast O(1) priority queue implementation 
 *  for the simulation event set problem." 
 *  Comm. of ACM, 31(10):1220-1227, Oct 1988
 */

#define CALENDAR_HASH(t) ((int)fmod((t)/width_, nbuckets_))

static class CalendarSchedulerClass : public TclClass {
public:
    CalendarSchedulerClass() : TclClass("Scheduler/Calendar") {}
    TclObject* create(int /* argc */, const char*const* /* argv */) {
        return (new CalendarScheduler);
    }
} class_calendar_sched;

CalendarScheduler::CalendarScheduler() : cal_clock_(clock_) {
    reinit(4, 1.0, cal_clock_);
}

CalendarScheduler::~CalendarScheduler() {
    // XXX free events?
    delete [] buckets_;
    qsize_ = 0;
    stat_qsize_ = 0;
}

void 
CalendarScheduler::insert(Event* e)
{
    int i;
    if (cal_clock_ > e->time_) {
        // may happen in RT scheduler
        cal_clock_ = e->time_;
        i = lastbucket_ = CALENDAR_HASH(cal_clock_);
    } else
        i = CALENDAR_HASH(e->time_);

    Event *head = buckets_[i].list_;
    Event *before=0;

    if (!head) {
        buckets_[i].list_ = e;
        e->next_ = e->prev_ = e;
        ++stat_qsize_; 
        ++buckets_[i].count_;
    } else {
        bool newhead;
        if (e->time_ >= head->prev_->time_) {
            // insert at the tail
            before = head;
            newhead = false;
        } else {
            // insert event in time sorted order, FIFO for sim-time events
            for (before = head; e->time_ >= before->time_; before = before->next_)
                ;
            newhead = (before == head);
        }

        e->next_ = before;
        e->prev_ = before->prev_;
        before->prev_ = e;
        e->prev_->next_ = e;
        if (newhead) {
            buckets_[i].list_ = e;
            //assert(e->time_ <= e->next_->time_);
        }
        //assert(e->prev_ != e);
        if (e->prev_->time_ != e->time_) {
            // unique timing
            ++stat_qsize_; 
            ++buckets_[i].count_;
        }
    }
    ++qsize_;
    //assert(e == buckets_[i].list_ ||  e->prev_->time_ <= e->time_);
    //assert(e == buckets_[i].list_->prev_ || e->next_->time_ >= e->time_);

    if (stat_qsize_ > top_threshold_) {
        resize(nbuckets_ << 1, cal_clock_);
        return;
    }
}

void 
CalendarScheduler::insert2(Event* e)
{
    // Same as insert, but for inserts e *before* any same-time-events, if
    //   there should be any.  Since it is used only by CalendarScheduler::newwidth(),
    //   some important checks present in insert() need not be performed.

    int i = CALENDAR_HASH(e->time_);
    Event *head = buckets_[i].list_;
    Event *before=0;
    if (!head) {
        buckets_[i].list_ = e;
        e->next_ = e->prev_ = e;
        ++stat_qsize_; 
        ++buckets_[i].count_;
    } else {
        bool newhead;
        if (e->time_ > head->prev_->time_) { //strict LIFO, so > and not >=
            // insert at the tail
            before = head;
            newhead = false;
        } else {
            // insert event in time sorted order, LIFO for sim-time events
            for (before = head; e->time_ > before->time_; before = before->next_)
                ;
            newhead = (before == head);
        }

        e->next_ = before;
        e->prev_ = before->prev_;
        before->prev_ = e;
        e->prev_->next_ = e;
        if (newhead) {
            buckets_[i].list_ = e;
            //assert(e->time_ <= e->next_->time_);
        }

        if (e != e->next_ && e->next_->time_ != e->time_) {
            // unique timing
            ++stat_qsize_; 
            ++buckets_[i].count_;
        }
    }
    //assert(e == buckets_[i].list_ ||  e->prev_->time_ <= e->time_);
    //assert(e == buckets_[i].list_->prev_ || e->next_->time_ >= e->time_);

    ++qsize_;
    // no need to check resizing
}

const Event*
CalendarScheduler::head()
{
    if (qsize_ == 0)
        return NULL;

    int l = -1, i = lastbucket_;
    int lastbucket_dec = (lastbucket_) ? lastbucket_ - 1 : nbuckets_ - 1;
    double diff;
    Event *e, *min_e = NULL;
#define CAL_DEQUEUE(x)                      \
do {                                \
    if ((e = buckets_[i].list_) != NULL) {          \
        diff = e->time_ - cal_clock_;           \
        if (diff < diff##x##_)  {           \
            l = i;                  \
            goto found_l;               \
        }                       \
        if (min_e == NULL || min_e->time_ > e->time_) { \
            min_e = e;              \
            l = i;                  \
        }                       \
    }                           \
    if (++i == nbuckets_) i = 0;                \
} while (0)
         
    // CAL_DEQUEUE applied successively will find the event to
    // dequeue (within one year) and keep track of the
    // minimum-priority event seen so far; the argument controls
    // the criteria used to decide whether the event should be
    // considered `within one year'.  Importantly, the number of
    // buckets should not be less than 4.
    CAL_DEQUEUE(0); 
    CAL_DEQUEUE(1); 
    for (; i != lastbucket_dec; ) {
        CAL_DEQUEUE(2);
    }
    // one last bucket is left unchecked - take the minimum
    // [could have done CAL_DEQUEUE(3) with diff3_ = bwidth*(nbuck*3/2-1)]
    e = buckets_[i].list_;
    if (min_e != NULL && 
        (e == NULL || min_e->time_ < e->time_))
        e = min_e;
    else {
        //assert(e);
        l = i;
    }
 found_l:
    //assert(buckets_[l].count_ >= 0);
    //assert(buckets_[l].list_ == e);

    /* l is the index of the bucket to dequeue, e is the event */
    /* 
     * still want to advance lastbucket_ and cal_clock_ to save
     * time when deque() follows. 
     */
    assert (l != -1);
    lastbucket_ = l;
    cal_clock_  = e->time_;
    
    return e;
}

Event* 
CalendarScheduler::deque()
{
    Event *e = const_cast<Event *>(head());

    if (!e)
        return 0;

    int l = lastbucket_;

    // update stats and unlink
    if (e->next_ == e || e->next_->time_ != e->time_) {
        --stat_qsize_;
        //assert(stat_qsize_ >= 0);
        --buckets_[l].count_;
        //assert(buckets_[l].count_ >= 0);

    }
    --qsize_;

    if (e->next_ == e)
        buckets_[l].list_ = 0;
    else {
        e->next_->prev_ = e->prev_;
        e->prev_->next_ = e->next_;
        buckets_[l].list_ = e->next_;
    }

    e->next_ = e->prev_ = NULL;


    //if (buckets_[l].count_ == 0)
    //  assert(buckets_[l].list_ == 0);

    if (stat_qsize_ < bot_threshold_) {
        resize(nbuckets_ >> 1, cal_clock_);
    }

    return e;
}

void 
CalendarScheduler::reinit(int nbuck, double bwidth, double start)
{
    buckets_ = new Bucket[nbuck];

    memset(buckets_, 0, sizeof(Bucket)*nbuck); //faster than ctor

    width_ = bwidth;
    nbuckets_ = nbuck;
    qsize_ = 0;
    stat_qsize_ = 0;

    lastbucket_ =  CALENDAR_HASH(start);

    diff0_ = bwidth*nbuck/2;
    diff1_ = diff0_ + bwidth;
    diff2_ = bwidth*nbuck;
    //diff3_ = bwidth*(nbuck*3/2-1);

    bot_threshold_ = (nbuck >> 1) - 2;
    top_threshold_ = (nbuck << 1);
}

void 
CalendarScheduler::resize(int newsize, double start)
{
    double bwidth = newwidth(newsize);

    if (newsize < 4)
        newsize = 4;

    Bucket *oldb = buckets_;
    int oldn = nbuckets_;

    reinit(newsize, bwidth, start);

    // copy events to new buckets
    int i;

    for (i = 0; i < oldn; ++i) {
        // we can do inserts faster, if we use insert2, but to
        // preserve FIFO, we have to start from the end of
        // each bucket and use insert2
        if  (oldb[i].list_) {
            Event *tail = oldb[i].list_->prev_;
            Event *e = tail; 
            do {
                Event* ep = e->prev_;
                e->next_ = e->prev_ = 0;
                insert2(e);
                e = ep;
            } while (e != tail);
        }
    }
    delete [] oldb;
}

// take samples from the most populated bucket.
double
CalendarScheduler::newwidth(int newsize)
{
    int i;
    int max_bucket = 0; // index of the fullest bucket
    for (i = 1; i < nbuckets_; ++i) {
        if (buckets_[i].count_ > buckets_[max_bucket].count_)
            max_bucket = i;
    }
    int nsamples = buckets_[max_bucket].count_;

    if (nsamples <= 4) return width_;
    
    double nw = buckets_[max_bucket].list_->prev_->time_ 
        - buckets_[max_bucket].list_->time_;
    assert(nw > 0);
    
    nw /= ((newsize < nsamples) ? newsize : nsamples); // min (newsize, nsamples)
    nw *= 4.0;
    
    return nw;
}

/*
 * Cancel an event.  It is an error to call this routine
 * when the event is not actually in the queue.  The caller
 * must free the event if necessary; this routine only removes
 * it from the scheduler queue.
 *
 */
void 
CalendarScheduler::cancel(Event* e)
{
    if (e->uid_ <= 0)   // event not in queue
        return;

    int i = CALENDAR_HASH(e->time_);

    assert(e->prev_->next_ == e);
    assert(e->next_->prev_ == e);

    if (e->next_ == e || 
        (e->next_->time_ != e->time_ &&
        (e->prev_->time_ != e->time_))) { 
        --stat_qsize_;
        assert(stat_qsize_ >= 0);
        --buckets_[i].count_;
        assert(buckets_[i].count_ >= 0);
    }

    if (e->next_ == e) {
        assert(buckets_[i].list_ == e);
        buckets_[i].list_ = 0;
    } else {
        e->next_->prev_ = e->prev_;
        e->prev_->next_ = e->next_;
        if (buckets_[i].list_ == e)
            buckets_[i].list_ = e->next_;
    }

    if (buckets_[i].count_ == 0)
        assert(buckets_[i].list_ == 0);

    e->uid_ = -e->uid_;
    e->next_ = e->prev_ = NULL;

    --qsize_;

    return;
}

Event * 
CalendarScheduler::lookup(scheduler_uid_t uid)
{
    for (int i = 0; i < nbuckets_; i++) {
        Event* head =  buckets_[i].list_;
        Event* p = head;
        if (p) {
            do {
                if (p->uid_== uid) 
                    return p;
                p = p->next_;
            } while (p != head);
        }
    }
    return NULL;
}

#ifndef WIN32
#include <sys/time.h>
#endif

class RealTimeScheduler : public CalendarScheduler {
public:
    RealTimeScheduler();
    virtual void run();
    double start() const { return start_; }
    virtual void reset();
protected:
    void sync() { clock_ = tod(); }
    double tod();
    double slop_;   // allowed drift between real-time and virt time
    double start_;  // starting time
};

static class RealTimeSchedulerClass : public TclClass {
public:
    RealTimeSchedulerClass() : TclClass("Scheduler/RealTime") {}
    TclObject* create(int /* argc */, const char*const* /* argv */) {
        return (new RealTimeScheduler);
    }
} class_realtime_sched;

RealTimeScheduler::RealTimeScheduler() : start_(0.0)
{
    bind("maxslop_", &slop_);
}

double
RealTimeScheduler::tod()
{
    timeval tv;
    gettimeofday(&tv, 0);
    double s = tv.tv_sec;
    s += (1e-6 * tv.tv_usec);
    return (s - start_);
}

void
RealTimeScheduler::reset()
{
    clock_ = SCHED_START;
    start_ = tod();
}

void 
RealTimeScheduler::run()
{ 
    static const double RTSCHEDULER_MINWAIT = 1.0e-3; // don't wait for less
    const Event *p;

    /*XXX*/
    instance_ = this;

    while (!halted_) {
        clock_ = tod();
        p = head();
        if (p && (clock_ - p->time_) > slop_) {
            fprintf(stderr,
                "RealTimeScheduler: warning: slop "
                "%f exceeded limit %f [clock_:%f, p->time_:%f]\n",
                clock_ - p->time_, slop_, clock_, p->time_);
        }
        // handle "old events"
        while (p && p->time_ <= clock_) {

            dispatch(deque(), clock_);
            if (halted_)
                return;
            p = head();
            clock_ = tod();
        }
        
        if (!p) {
            // blocking wait for TCL events
            Tcl_WaitForEvent(0); // no sim events, wait forever
            clock_ = tod();
        } else {
            double diff = p->time_ - clock_;
            // blocking wait only if there is enough time
            if (diff > RTSCHEDULER_MINWAIT) {
                Tcl_Time to;
                to.sec = long(diff);
                to.usec = long(1e6*(diff - to.sec));
                Tcl_WaitForEvent(&to);    // block
                clock_ = tod();
            }
        }
        Tcl_DoOneEvent(TCL_DONT_WAIT);
    }
    // we reach here only if halted
}

---