diff --git a/kernel/include/profile.h b/kernel/include/profile.h
index 5d1cd2ca..fbd3cfa9 100644
--- a/kernel/include/profile.h
+++ b/kernel/include/profile.h
@@ -10,6 +10,7 @@
 #define PROFILE_EVENT_MIN            PROFILE_OFFLOAD_MAX
 #define __NR_profile                                 701
 
+#define PROF_JOB                       0x40000000
 #define PROF_PROC                      0x80000000
 #define PROF_CLEAR                           0x01
 #define PROF_ON                              0x02
@@ -21,13 +22,13 @@ struct profile_event {
 	uint64_t tsc;
 };
 
-/* 
+/*
  * The layout of profile events is as follows:
  * [0,PROFILE_SYSCALL_MAX) - syscalls
  * [PROFILE_SYSCALL_MAX,PROFILE_OFFLOAD_MAX) - syscall offloads
  * [PROFILE_OFFLOAD_MAX,PROFILE_EVENT_MAX) - general events
  *
- * XXX: Make sure to fill in prof_event_names in profile.c 
+ * XXX: Make sure to fill in prof_event_names in profile.c
  * for each added profiled event.
  */
 enum profile_event_type {
@@ -43,7 +44,9 @@ enum profile_event_type profile_syscall2offload(enum profile_event_type sc);
 void profile_event_add(enum profile_event_type type, uint64_t tsc);
 void profile_print_thread_stats(struct thread *thread);
 void profile_print_proc_stats(struct process *proc);
+void profile_print_job_stats(struct process *proc);
 void profile_accumulate_events(struct thread *thread, struct process *proc);
+int profile_accumulate_and_print_job_events(struct process *proc);
 int profile_alloc_events(struct thread *thread);
 void profile_dealloc_thread_events(struct thread *thread);
 void profile_dealloc_proc_events(struct process *proc);
diff --git a/kernel/process.c b/kernel/process.c
index c8a2e363..97b21fe4 100644
--- a/kernel/process.c
+++ b/kernel/process.c
@@ -2218,7 +2218,12 @@ release_process(struct process *proc)
 
 	if (proc->tids) kfree(proc->tids);
 #ifdef PROFILE_ENABLE
-	profile_print_proc_stats(proc);
+	if (proc->nr_processes) {
+		profile_accumulate_and_print_job_events(proc);
+	}
+	else {
+		profile_print_proc_stats(proc);
+	}
 	profile_dealloc_proc_events(proc);
 #endif // PROFILE_ENABLE
 	kfree(proc);
diff --git a/kernel/profile.c b/kernel/profile.c
index 0472d01c..7025de6b 100644
--- a/kernel/profile.c
+++ b/kernel/profile.c
@@ -51,6 +51,13 @@ char *profile_event_names[] =
 	""
 };
 
+mcs_lock_node_t job_profile_lock = {0, NULL};
+struct profile_event *job_profile_events = NULL;
+int job_nr_processes = -1;
+int job_nr_processes_left = -1;
+
+
+
 enum profile_event_type profile_syscall2offload(enum profile_event_type sc)
 {
 	return (PROFILE_SYSCALL_MAX + sc);
@@ -85,6 +92,8 @@ void profile_print_thread_stats(struct thread *thread)
 	int i;
 	unsigned long flags;
 
+	if (!thread->profile_events) return;
+
 	flags = kprintf_lock();
 
 	for (i = 0; i < PROFILE_SYSCALL_MAX; ++i) {
@@ -134,6 +143,8 @@ void profile_print_proc_stats(struct process *proc)
 	int i;
 	unsigned long flags;
 
+	if (!proc->profile_events) return;
+
 	flags = kprintf_lock();
 
 	for (i = 0; i < PROFILE_SYSCALL_MAX; ++i) {
@@ -177,17 +188,124 @@ void profile_print_proc_stats(struct process *proc)
 	kprintf_unlock(flags);
 }
 
+int profile_accumulate_and_print_job_events(struct process *proc)
+{
+	int i;
+	unsigned long flags;
+	struct mcs_lock_node mcs_node;
+
+	mcs_lock_lock(&job_profile_lock, &mcs_node);
+
+	/* First process? */
+	if (job_nr_processes == -1) {
+		job_nr_processes = proc->nr_processes;
+		job_nr_processes_left = proc->nr_processes;
+	}
+
+	--job_nr_processes_left;
+
+	/* Allocate event counters */
+	if (!job_profile_events) {
+
+		job_profile_events = kmalloc(sizeof(*job_profile_events) *
+				PROFILE_EVENT_MAX, IHK_MC_AP_NOWAIT);
+
+		if (!job_profile_events) {
+			kprintf("%s: ERROR: allocating job profile counters\n",
+					__FUNCTION__);
+			return -ENOMEM;
+		}
+
+		memset(job_profile_events, 0,
+				sizeof(*job_profile_events) * PROFILE_EVENT_MAX);
+	}
+
+	/* Accumulate process */
+	for (i = 0; i < PROFILE_EVENT_MAX; ++i) {
+		if (!proc->profile_events[i].tsc)
+			continue;
+
+		job_profile_events[i].tsc += proc->profile_events[i].tsc;
+		job_profile_events[i].cnt += proc->profile_events[i].cnt;
+		proc->profile_events[i].tsc = 0;
+		proc->profile_events[i].cnt = 0;
+	}
+
+	/* Last process? */
+	if (job_nr_processes_left == 0) {
+		flags = kprintf_lock();
+
+		for (i = 0; i < PROFILE_SYSCALL_MAX; ++i) {
+			if (!job_profile_events[i].cnt &&
+					!job_profile_events[i + PROFILE_SYSCALL_MAX].cnt)
+				continue;
+
+			__kprintf("JOB: (%2d) (%3d,%20s): %6u %6lukC offl: %6u %6lukC\n",
+					job_nr_processes,
+					i,
+					syscall_name[i],
+					job_profile_events[i].cnt,
+					(job_profile_events[i].tsc /
+					 (job_profile_events[i].cnt ?
+					  job_profile_events[i].cnt : 1))
+					/ 1000,
+					job_profile_events[i + PROFILE_SYSCALL_MAX].cnt,
+					(job_profile_events[i + PROFILE_SYSCALL_MAX].tsc /
+					 (job_profile_events[i + PROFILE_SYSCALL_MAX].cnt ?
+					  job_profile_events[i + PROFILE_SYSCALL_MAX].cnt : 1))
+					/ 1000
+					);
+
+			job_profile_events[i].tsc = 0;
+			job_profile_events[i].cnt = 0;
+		}
+
+		for (i = PROFILE_EVENT_MIN; i < PROFILE_EVENT_MAX; ++i) {
+
+			if (!job_profile_events[i].cnt)
+				continue;
+
+			__kprintf("JOB: (%2d) (%3d,%20s): %6u %6lukC \n",
+					job_nr_processes,
+					i,
+					profile_event_names[i - PROFILE_EVENT_MIN],
+					job_profile_events[i].cnt,
+					(job_profile_events[i].tsc /
+					 (job_profile_events[i].cnt ?
+					  job_profile_events[i].cnt : 1))
+					/ 1000);
+
+			job_profile_events[i].tsc = 0;
+			job_profile_events[i].cnt = 0;
+		}
+
+		kprintf_unlock(flags);
+
+		/* Reset job process indicators */
+		job_nr_processes = -1;
+		job_nr_processes_left = -1;
+	}
+
+	mcs_lock_unlock(&job_profile_lock, &mcs_node);
+
+	return 0;
+}
+
 void profile_accumulate_events(struct thread *thread,
 		struct process *proc)
 {
 	int i;
 	struct mcs_lock_node mcs_node;
 
+	if (!thread->profile_events || !proc->profile_events) return;
+
 	mcs_lock_lock(&proc->profile_lock, &mcs_node);
 
 	for (i = 0; i < PROFILE_EVENT_MAX; ++i) {
 		proc->profile_events[i].tsc += thread->profile_events[i].tsc;
 		proc->profile_events[i].cnt += thread->profile_events[i].cnt;
+		thread->profile_events[i].tsc = 0;
+		thread->profile_events[i].cnt = 0;
 	}
 
 	mcs_lock_unlock(&proc->profile_lock, &mcs_node);
@@ -198,17 +316,19 @@ int profile_alloc_events(struct thread *thread)
 	struct process *proc = thread->proc;
 	struct mcs_lock_node mcs_node;
 
-	thread->profile_events = kmalloc(sizeof(*thread->profile_events) *
-			PROFILE_EVENT_MAX, IHK_MC_AP_NOWAIT);
-
 	if (!thread->profile_events) {
-		kprintf("%s: ERROR: allocating thread private profile counters\n",
-				__FUNCTION__);
-		return -ENOMEM;
-	}
+		thread->profile_events = kmalloc(sizeof(*thread->profile_events) *
+				PROFILE_EVENT_MAX, IHK_MC_AP_NOWAIT);
 
-	memset(thread->profile_events, 0,
-			sizeof(*thread->profile_events) * PROFILE_EVENT_MAX);
+		if (!thread->profile_events) {
+			kprintf("%s: ERROR: allocating thread private profile counters\n",
+					__FUNCTION__);
+			return -ENOMEM;
+		}
+
+		memset(thread->profile_events, 0,
+				sizeof(*thread->profile_events) * PROFILE_EVENT_MAX);
+	}
 
 	mcs_lock_lock(&proc->profile_lock, &mcs_node);
 	if (!proc->profile_events) {
@@ -218,6 +338,7 @@ int profile_alloc_events(struct thread *thread)
 		if (!proc->profile_events) {
 			kprintf("%s: ERROR: allocating proc private profile counters\n",
 					__FUNCTION__);
+			mcs_lock_unlock(&proc->profile_lock, &mcs_node);
 			return -ENOMEM;
 		}
 
@@ -242,6 +363,8 @@ void profile_dealloc_proc_events(struct process *proc)
 
 void static profile_clear_thread(struct thread *thread)
 {
+	if (!thread->profile_events) return;
+
 	memset(thread->profile_events, 0,
 			sizeof(*thread->profile_events) * PROFILE_EVENT_MAX);
 }
@@ -249,18 +372,56 @@ void static profile_clear_thread(struct thread *thread)
 int do_profile(int flag)
 {
 	struct thread *thread = cpu_local_var(current);
+	struct process *proc = thread->proc;
 
-	/* Process level? */
-	if (flag & PROF_PROC) {
+	/* Job level? */
+	if (flag & PROF_JOB) {
 		if (flag & PROF_PRINT) {
-			profile_print_proc_stats(thread->proc);
+			struct mcs_rwlock_node lock;
+			struct thread *_thread;
+
+			/* Accumulate events from all threads to process level */
+			mcs_rwlock_reader_lock_noirq(&proc->threads_lock, &lock);
+			list_for_each_entry(_thread, &proc->threads_list,
+					siblings_list) {
+				profile_accumulate_events(_thread, proc);
+			}
+			mcs_rwlock_reader_unlock_noirq(&proc->threads_lock, &lock);
+
+			/* Accumulate events to job level */
+			return profile_accumulate_and_print_job_events(proc);
+		}
+	}
+	/* Process level? */
+	else if (flag & PROF_PROC) {
+		struct mcs_rwlock_node lock;
+		struct thread *_thread;
+
+		/* Accumulate events from all threads */
+		mcs_rwlock_reader_lock_noirq(&proc->threads_lock, &lock);
+
+		list_for_each_entry(_thread, &proc->threads_list,
+				siblings_list) {
+			if (flag & PROF_PRINT) {
+				profile_accumulate_events(_thread, proc);
+			}
+
+			if (flag & PROF_CLEAR) {
+				profile_clear_thread(_thread);
+			}
+
+			if (flag & PROF_ON) {
+				_thread->profile = 1;
+			}
+			else if (flag & PROF_OFF) {
+				_thread->profile = 0;
+			}
 		}
 
-		if (flag & PROF_ON) {
-			thread->profile = 1;
-		}
-		else if (flag & PROF_OFF) {
-			thread->profile = 0;
+		mcs_rwlock_reader_unlock_noirq(&proc->threads_lock, &lock);
+
+		if (flag & PROF_PRINT) {
+			profile_print_proc_stats(proc);
 		}
 	}
 	/* Thread level */