--- linux-2.6.0-test11-ht1/kernel/sched.c	2003-12-05 13:32:59.000000000 +1100
+++ linux-2.6.0-test11-b1/kernel/sched.c	2003-12-06 00:34:12.832783883 +1100
@@ -141,7 +141,7 @@
 		INTERACTIVE_DELTA)
 
 #define TASK_INTERACTIVE(p) \
-	((p)->prio <= (p)->static_prio - DELTA(p))
+	((p)->prio <= (p)->static_prio - DELTA(p) && !batch_task(p))
 
 #define JUST_INTERACTIVE_SLEEP(p) \
 	(JIFFIES_TO_NS(MAX_SLEEP_AVG * \
@@ -169,6 +169,13 @@
 
 static inline unsigned int task_timeslice(task_t *p)
 {
+	/*
+	 * Batch tasks get much longer timeslices to optimise cpu throughput.
+	 * Since they yield to any other tasks this is not a problem.
+	 */
+	if (unlikely(batch_task(p)))
+		return BASE_TIMESLICE(p) * 10;
+
 	return BASE_TIMESLICE(p);
 }
 
@@ -238,9 +245,9 @@ typedef struct cpu_s {
 struct runqueue {
 	spinlock_t lock;
 	unsigned long nr_running, nr_switches, expired_timestamp,
-			nr_uninterruptible;
+			nr_uninterruptible, nr_batch;
 	struct mm_struct *prev_mm;
-	prio_array_t *active, *expired, arrays[2];
+	prio_array_t *active, *expired, *batch, arrays[3];
 	int prev_cpu_load[NR_CPUS];
 #ifdef CONFIG_NUMA
 	atomic_t *node_nr_running;
@@ -319,9 +326,11 @@ __init void node_nr_running_init(void)
 
 #else /* !CONFIG_NUMA */
 
-# define nr_running_init(rq)   do { } while (0)
-# define nr_running_inc(rq)    do { (rq)->nr_running++; } while (0)
-# define nr_running_dec(rq)    do { (rq)->nr_running--; } while (0)
+# define nr_running_init(rq)	do { } while (0)
+# define nr_running_inc(rq)	do { (rq)->nr_running++; } while (0)
+# define nr_running_dec(rq)	do { (rq)->nr_running--; } while (0)
+# define nr_batch_inc(rq)	do { (rq)->nr_batch++; } while (0)
+# define nr_batch_dec(rq)	do { (rq)->nr_batch--; } while (0)
 
 #endif /* CONFIG_NUMA */
 
@@ -414,7 +423,7 @@ static int effective_prio(task_t *p)
 	prio = p->static_prio - bonus;
 	if (prio < MAX_RT_PRIO)
 		prio = MAX_RT_PRIO;
-	if (prio > MAX_PRIO-1)
+	if (prio > MAX_PRIO-1 || batch_task(p))
 		prio = MAX_PRIO-1;
 	return prio;
 }
@@ -424,7 +433,12 @@ static int effective_prio(task_t *p)
  */
 static inline void __activate_task(task_t *p, runqueue_t *rq)
 {
-	enqueue_task(p, rq->active);
+	if (unlikely(batch_task(p))){
+		enqueue_task(p, rq->batch);
+		nr_batch_inc(rq);
+	} else
+		enqueue_task(p, rq->active);
+
 	nr_running_inc(rq);
 }
 
@@ -548,6 +562,8 @@ static inline void activate_task(task_t 
 static inline void deactivate_task(struct task_struct *p, runqueue_t *rq)
 {
 	nr_running_dec(rq);
+	if (unlikely(batch_task(p)))
+		nr_batch_dec(rq);
 	if (p->state == TASK_UNINTERRUPTIBLE)
 		rq->nr_uninterruptible++;
 	dequeue_task(p, p->array);
@@ -675,16 +691,22 @@ static void wake_up_cpu(runqueue_t *rq, 
 			return resched_cpu(curr_cpu->cpu);
 	}
 
-	if (p->prio < cpu_curr_ptr(cpu)->prio)
-		return resched_task(cpu_curr_ptr(cpu));
+	/*
+	 * SCHED_BATCH tasks never preempt but are always preempted by
+	 * other policy tasks
+	 */
+	if (!batch_task(p) && (p->prio < cpu_curr_ptr(cpu)->prio ||
+		batch_task(cpu_curr_ptr(cpu))))
+			return resched_task(cpu_curr_ptr(cpu));
 
 	for_each_sibling(idx, rq) {
 		curr_cpu = rq->cpu + idx;
 		if (!task_allowed(p, curr_cpu->cpu))
 			continue;
 		curr = curr_cpu->curr;
-		if (p->prio < curr->prio)
-			return resched_task(curr);
+		if (!batch_task(p) && (p->prio < curr->prio ||
+			batch_task(curr)))
+				return resched_task(curr);
 	}
 }
 
@@ -797,9 +819,20 @@ void wake_up_forked_process(task_t * p)
 		p->prio = current->prio;
 		list_add_tail(&p->run_list, &current->run_list);
 		p->array = current->array;
-		p->array->nr_active++;
+		if (unlikely(batch_task(p))){
+			p->array = task_rq(current)->batch;
+			p->array->nr_active++;
+			nr_batch_inc(rq);
+		} else {
+			p->array = current->array;
+			p->array->nr_active++;
+		}
 		nr_running_inc(rq);
 	}
+
+	if (unlikely(batch_task(current) && !batch_task(p)))
+		resched_task(current);
+
 	task_rq_unlock(rq, &flags);
 }
 
@@ -1228,7 +1261,13 @@ static inline void pull_task(runqueue_t 
 	nr_running_dec(src_rq);
 	set_task_cpu(p, this_cpu);
 	nr_running_inc(this_rq);
-	enqueue_task(p, this_rq->active);
+	if (unlikely(batch_task(p))){
+		nr_batch_dec(src_rq);
+		nr_batch_inc(this_rq);
+		enqueue_task(p, this_rq->batch);
+	} else
+		enqueue_task(p, this_rq->active);
+
 	wake_up_cpu(this_rq, this_cpu, p);
 }
 
@@ -1593,13 +1632,36 @@ void scheduler_tick(int user_ticks, int 
 			cpustat->iowait += sys_ticks;
 		else
 			cpustat->idle += sys_ticks;
+
+		/*
+		 * Check to see if batch tasks are waiting to run and
+		 * reschedule the idle task if no normal tasks are
+		 * on this runqueue.
+		 */
+		if (rq->nr_batch && rq->nr_batch == rq->nr_running) {
+			resched_task(p);
+			goto out;
+		}
 		rebalance_tick(rq, cpu, 1);
 		return;
 	}
-	if (TASK_NICE(p) > 0)
-		cpustat->nice += user_ticks;
-	else
-		cpustat->user += user_ticks;
+
+	if (unlikely(batch_task(p))){
+		/*
+		 * SCHED_BATCH tasks are counted as idle time.
+		 */
+		cpustat->idle += user_ticks;
+		if (rq->nr_running > rq->nr_batch){
+			resched_task(p);
+			goto out;
+		}
+	} else {
+		if (TASK_NICE(p) > 0)
+			cpustat->nice += user_ticks;
+		else
+			cpustat->user += user_ticks;
+	}
+
 	cpustat->system += sys_ticks;
 
 	/* Task might have expired already, but not scheduled off yet */
@@ -1772,16 +1834,40 @@ pick_idle:
 		/*
 		 * Switch the active and expired arrays.
 		 */
-		rq->active = rq->expired;
-		rq->expired = array;
+		if (likely(rq->expired->nr_active)){
+			rq->active = rq->expired;
+			rq->expired = array;
+			rq->expired_timestamp = 0;
+		} else {
+			/*
+			 * Switch to the batch array if there are no
+			 * normal tasks left waiting.
+			 */
+			rq->active = rq->batch;
+			rq->batch = array;
+		}
+
 		array = rq->active;
-		rq->expired_timestamp = 0;
 	}
 
 new_array:
 	idx = sched_find_first_bit(array->bitmap);
-  	queue = array->queue + idx;
-  	next = list_entry(queue->next, task_t, run_list);
+	queue = array->queue + idx;
+	next = list_entry(queue->next, task_t, run_list);
+
+	/*
+	 * If a batch process is on the active array and normal processes
+	 * are still running move the batch process to the batch
+	 * array.
+	 */
+	if (unlikely(batch_task(next))){
+		if (rq->nr_running > rq->nr_batch){
+			dequeue_task(next, array);
+			enqueue_task(next, rq->batch);
+			goto pick_next_task;
+		}
+	}
+
 	if ((next != prev) && (rq_nr_cpus(rq) > 1)) {
 		struct list_head *tmp = queue->next;
 
@@ -1793,6 +1879,10 @@ new_array:
 			}
 			idx = find_next_bit(array->bitmap, MAX_PRIO, ++idx);
 			if (idx == MAX_PRIO) {
+				/*
+				 * Batch tasks are not considered for siblings
+				 * on busy shared runqueues.
+				 */
 				if (retry || !rq->expired->nr_active) {
 					goto pick_idle;
 				}
@@ -2134,10 +2224,11 @@ void set_user_nice(task_t *p, long nice)
 		enqueue_task(p, array);
 		/*
 		 * If the task increased its priority or is running and
-		 * lowered its priority, then reschedule its CPU:
+		 * lowered its priority or is batch, then reschedule its CPU:
 		 */
-		if (delta < 0 || (delta > 0 && task_running(p)))
-			resched_task(cpu_curr_ptr(task_cpu(p)));
+		if (delta < 0 || ((delta > 0 || batch_task(p)) &&
+			task_running(p)))
+				resched_task(cpu_curr_ptr(task_cpu(p)));
 	}
 out_unlock:
 	task_rq_unlock(rq, &flags);
@@ -2275,8 +2366,8 @@ static int setscheduler(pid_t pid, int p
 	else {
 		retval = -EINVAL;
 		if (policy != SCHED_FIFO && policy != SCHED_RR &&
-				policy != SCHED_NORMAL)
-			goto out_unlock;
+			policy != SCHED_NORMAL && policy !=SCHED_BATCH)
+				goto out_unlock;
 	}
 
 	/*
@@ -2286,8 +2377,9 @@ static int setscheduler(pid_t pid, int p
 	retval = -EINVAL;
 	if (lp.sched_priority < 0 || lp.sched_priority > MAX_USER_RT_PRIO-1)
 		goto out_unlock;
-	if ((policy == SCHED_NORMAL) != (lp.sched_priority == 0))
-		goto out_unlock;
+	if ((policy == SCHED_NORMAL || policy == SCHED_BATCH) !=
+		(lp.sched_priority == 0))
+			goto out_unlock;
 
 	retval = -EPERM;
 	if ((policy == SCHED_FIFO || policy == SCHED_RR) &&
@@ -2308,7 +2400,7 @@ static int setscheduler(pid_t pid, int p
 	p->policy = policy;
 	p->rt_priority = lp.sched_priority;
 	oldprio = p->prio;
-	if (policy != SCHED_NORMAL)
+	if (policy != SCHED_NORMAL && policy != SCHED_BATCH)
 		p->prio = MAX_USER_RT_PRIO-1 - p->rt_priority;
 	else
 		p->prio = p->static_prio;
@@ -2320,10 +2412,11 @@ static int setscheduler(pid_t pid, int p
 		 * this runqueue and our priority is higher than the current's
 		 */
 		if (cpu_curr_ptr(task_cpu(p)) == p) {
-			if (p->prio > oldprio)
+			if (p->prio > oldprio || batch_task(p))
+				resched_task(cpu_curr_ptr(task_cpu(p)));
+		} else if (!batch_task(p) &&
+			p->prio < cpu_curr_ptr(task_cpu(p))->prio)
 				resched_task(cpu_curr_ptr(task_cpu(p)));
-		} else if (p->prio < cpu_curr_ptr(task_cpu(p))->prio)
-			resched_task(cpu_curr_ptr(task_cpu(p)));
 	}
 
 out_unlock:
@@ -2528,7 +2621,10 @@ asmlinkage long sys_sched_yield(void)
 	 */
 	if (likely(!rt_task(current))) {
 		dequeue_task(current, array);
-		enqueue_task(current, rq->expired);
+		if (unlikely(batch_task(current)))
+			enqueue_task(current, rq->batch);
+		else
+			enqueue_task(current, rq->expired);
 	} else {
 		list_del(&current->run_list);
 		list_add_tail(&current->run_list, array->queue + current->prio);
@@ -2613,6 +2709,7 @@ asmlinkage long sys_sched_get_priority_m
 		ret = MAX_USER_RT_PRIO-1;
 		break;
 	case SCHED_NORMAL:
+	case SCHED_BATCH:
 		ret = 0;
 		break;
 	}
@@ -2636,6 +2733,7 @@ asmlinkage long sys_sched_get_priority_m
 		ret = 1;
 		break;
 	case SCHED_NORMAL:
+	case SCHED_BATCH:
 		ret = 0;
 	}
 	return ret;
@@ -3080,6 +3178,7 @@ void __init sched_init(void)
 		rq = cpu_rq(i);
 		rq->active = rq->arrays;
 		rq->expired = rq->arrays + 1;
+		rq->batch = rq->arrays + 2;
 		spin_lock_init(&rq->lock);
 		INIT_LIST_HEAD(migration_queue(i));
 		rq->nr_cpus = 1;
@@ -3087,7 +3186,7 @@ void __init sched_init(void)
 		atomic_set(&rq->nr_iowait, 0);
 		nr_running_init(rq);
 
-		for (j = 0; j < 2; j++) {
+		for (j = 0; j < 3; j++) {
 			array = rq->arrays + j;
 			for (k = 0; k < MAX_PRIO; k++) {
 				INIT_LIST_HEAD(array->queue + k);
--- linux-2.6.0-test11-ht1/include/linux/sched.h	2003-12-05 13:27:40.000000000 +1100
+++ linux-2.6.0-test11-b1/include/linux/sched.h	2003-12-05 21:10:11.000000000 +1100
@@ -126,6 +126,7 @@ extern unsigned long nr_iowait(void);
 #define SCHED_NORMAL		0
 #define SCHED_FIFO		1
 #define SCHED_RR		2
+#define SCHED_BATCH		3
 
 struct sched_param {
 	int sched_priority;
@@ -303,6 +304,7 @@ struct signal_struct {
 #define MAX_PRIO		(MAX_RT_PRIO + 40)
 
 #define rt_task(p)		((p)->prio < MAX_RT_PRIO)
+#define batch_task(p)		((p)->policy == SCHED_BATCH)
 
 /*
  * Some day this will be a full-fledged user tracking system..