Introduce a better id allocation algorithm for PTabs

10 files changed, 352 insertions, 131 deletions

diff --git a/erts/doc/src/erl.xml b/erts/doc/src/erl.xml
index 9f5de8a9ae..a7b05899fd 100644
--- a/erts/doc/src/erl.xml
+++ b/erts/doc/src/erl.xml
@@ -637,10 +637,11 @@
 	<p>Se also <seealso marker="stdlib:io#printable_range/0">
 	io:printable_range/0</seealso>.</p>
       </item>
-      <tag><marker id="+P"/><marker id="max_processes"><c><![CDATA[+P Number]]></c></marker></tag>
+      <tag><marker id="+P"/><marker id="max_processes"><c><![CDATA[+P Number|legacy]]></c></marker></tag>
       <item>
 	<p>Sets the maximum number of simultaneously existing processes for this
-          system. Valid range for <c>Number</c> is <c>[1024-134217727]</c></p>
+          system if a <c>Number</c> is passed as value. Valid range for
+	<c>Number</c> is <c>[1024-134217727]</c></p>
 	<p><em>NOTE</em>: The actual maximum chosen may be much larger than
 	the <c>Number</c> passed. Currently the runtime system often,
 	but not always, chooses a value that is a power of 2. This might,
@@ -648,11 +649,19 @@
 	checked by calling
 	<seealso marker="erlang#system_info_process_limit">erlang:system_info(process_limit)</seealso>.</p>
 	<p>The default value is <c>262144</c></p>
+	<p>If <c>legacy</c> is passed as value, the legacy algorithm for
+	allocation of process identifiers will be used. Using the legacy
+	algorithm, identifiers will be allocated in a strictly increasing
+	fashion until largest possible identifier has been reached. Note that
+	this algorithm suffers from performance issues and can under certain
+	circumstances be extremely expensive. The legacy algoritm is deprecated,
+	and the <c>legacy</c> option is scheduled for removal in OTP-R18.</p>
       </item>
-      <tag><marker id="+Q"/><marker id="max_ports"><c><![CDATA[+Q Number]]></c></marker></tag>
+      <tag><marker id="+Q"/><marker id="max_ports"><c><![CDATA[+Q Number|legacy]]></c></marker></tag>
       <item>
 	<p>Sets the maximum number of simultaneously existing ports for this
-          system. Valid range for <c>Number</c> is <c>[1024-134217727]</c></p>
+          system if a Number is passed as value. Valid range for <c>Number</c>
+	is <c>[1024-134217727]</c></p>
 	<p><em>NOTE</em>: The actual maximum chosen may be much larger than
 	the actual <c>Number</c> passed. Currently the runtime system often,
 	but not always, chooses a value that is a power of 2. This might,
@@ -671,6 +680,13 @@
 	for setting the maximum number of simultaneously existing ports. This
 	environment variable is deprecated, and scheduled for removal in
 	OTP-R17, but can still be used.</p>
+	<p>If <c>legacy</c> is passed as value, the legacy algorithm for
+	allocation of port identifiers will be used. Using the legacy
+	algorithm, identifiers will be allocated in a strictly increasing
+	fashion until largest possible identifier has been reached. Note that
+	this algorithm suffers from performance issues and can under certain
+	circumstances be extremely expensive. The legacy algoritm is deprecated,
+	and the <c>legacy</c> option is scheduled for removal in OTP-R18.</p>
       </item>
       <tag><marker id="compat_rel"><c><![CDATA[+R ReleaseNumber]]></c></marker></tag>
       <item>
diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c
index d748f86d75..700ca127e6 100644
--- a/erts/emulator/beam/erl_alloc.c
+++ b/erts/emulator/beam/erl_alloc.c
@@ -2142,7 +2142,6 @@ erts_memory(int *print_to_p, void *print_to_arg, void *proc, Eterm earg)
 
 
     if (want_tot_or_sys || want.processes || want.processes_used) {
-	int max_processes = erts_ptab_max(&erts_proc);
 	UWord tmp;
 
 	if (ERTS_MEM_NEED_ALL_ALCU)
@@ -2152,7 +2151,7 @@ erts_memory(int *print_to_p, void *print_to_arg, void *proc, Eterm earg)
 		      fi, ERTS_ALC_NO_FIXED_SIZES);
 	    tmp = alcu_size(ERTS_ALC_A_EHEAP, NULL, 0);
 	}
-	tmp += max_processes*sizeof(erts_smp_atomic_t);
+	tmp += erts_ptab_mem_size(&erts_proc);
 	tmp += erts_bif_timer_memory_size();
 	tmp += erts_tot_link_lh_size();
 
@@ -2278,13 +2277,11 @@ struct aa_values {
 Eterm
 erts_allocated_areas(int *print_to_p, void *print_to_arg, void *proc)
 {
-#define MAX_AA_VALUES (23)
+#define MAX_AA_VALUES (24)
     struct aa_values values[MAX_AA_VALUES];
     Eterm res = THE_NON_VALUE;
     int i, length;
     Uint reserved_atom_space, atom_space;
-    int max_processes = erts_ptab_max(&erts_proc);
-    int max_ports = erts_ptab_max(&erts_port);
 
     if (proc) {
 	ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN
@@ -2315,8 +2312,8 @@ erts_allocated_areas(int *print_to_p, void *print_to_arg, void *proc)
 
     values[i].arity = 2;
     values[i].name = "static";
-    values[i].ui[0] = 
-	max_ports*sizeof(erts_smp_atomic_t)	/* Port table */
+    values[i].ui[0] =
+	sizeof(ErtsPTab)*2			/* proc & port tables */
 	+ erts_timer_wheel_memory_size();	/* Timer wheel */
     i++;
 
@@ -2395,7 +2392,12 @@ erts_allocated_areas(int *print_to_p, void *print_to_arg, void *proc)
 
     values[i].arity = 2;
     values[i].name = "process_table";
-    values[i].ui[0] = max_processes*sizeof(Process*);
+    values[i].ui[0] = erts_ptab_mem_size(&erts_proc);
+    i++;
+
+    values[i].arity = 2;
+    values[i].name = "port_table";
+    values[i].ui[0] = erts_ptab_mem_size(&erts_port);
     i++;
 
     values[i].arity = 2;
diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c
index e6a96d427f..b5bac2151d 100644
--- a/erts/emulator/beam/erl_init.c
+++ b/erts/emulator/beam/erl_init.c
@@ -131,8 +131,10 @@ extern void ConWaitForExit(void);
 
 static void erl_init(int ncpu,
 		     int proc_tab_sz,
+		     int legacy_proc_tab,
 		     int port_tab_sz,
-		     int port_tab_sz_ignore_files);
+		     int port_tab_sz_ignore_files,
+		     int legacy_port_tab);
 
 static erts_atomic_t exiting;
 
@@ -280,7 +282,9 @@ erts_short_init(void)
     int ncpu = early_init(NULL, NULL);
     erl_init(ncpu,
 	     ERTS_DEFAULT_MAX_PROCESSES,
+	     0,
 	     ERTS_DEFAULT_MAX_PORTS,
+	     0,
 	     0);
     erts_initialized = 1;
 }
@@ -288,8 +292,10 @@ erts_short_init(void)
 static void
 erl_init(int ncpu,
 	 int proc_tab_sz,
+	 int legacy_proc_tab,
 	 int port_tab_sz,
-	 int port_tab_sz_ignore_files)
+	 int port_tab_sz_ignore_files,
+	 int legacy_port_tab)
 {
     init_benchmarking();
 
@@ -297,7 +303,7 @@ erl_init(int ncpu,
     erts_init_gc();
     erts_init_time();
     erts_init_sys_common_misc();
-    erts_init_process(ncpu, proc_tab_sz);
+    erts_init_process(ncpu, proc_tab_sz, legacy_proc_tab);
     erts_init_scheduling(no_schedulers,
 			 no_schedulers_online);
     erts_init_cpu_topology(); /* Must be after init_scheduling */
@@ -327,7 +333,7 @@ erl_init(int ncpu,
     init_dist();
     erl_drv_thr_init();
     erts_init_async();
-    erts_init_io(port_tab_sz, port_tab_sz_ignore_files);
+    erts_init_io(port_tab_sz, port_tab_sz_ignore_files, legacy_port_tab);
     init_load();
     erts_init_bif();
     erts_init_bif_chksum();
@@ -923,6 +929,9 @@ erl_start(int argc, char **argv)
     int proc_tab_sz = ERTS_DEFAULT_MAX_PROCESSES;
     int port_tab_sz = ERTS_DEFAULT_MAX_PORTS;
     int port_tab_sz_ignore_files = 0;
+    int legacy_proc_tab = 0;
+    int legacy_port_tab = 0;
+
 
     envbufsz = sizeof(envbuf);
     if (erts_sys_getenv_raw(ERL_MAX_ETS_TABLES_ENV, envbuf, &envbufsz) == 0)
@@ -1268,27 +1277,35 @@ erl_start(int argc, char **argv)
 
 	case 'P': /* set maximum number of processes */
 	    arg = get_arg(argv[i]+2, argv[i+1], &i);
-	    errno = 0;
-	    proc_tab_sz = strtol(arg, NULL, 10);
-	    if (errno != 0
-		|| proc_tab_sz < ERTS_MIN_PROCESSES
-		|| ERTS_MAX_PROCESSES < proc_tab_sz) {
-		erts_fprintf(stderr, "bad number of processes %s\n", arg);
-		erts_usage();
+	    if (strcmp(arg, "legacy") == 0)
+		legacy_proc_tab = 1;
+	    else {
+		errno = 0;
+		proc_tab_sz = strtol(arg, NULL, 10);
+		if (errno != 0
+		    || proc_tab_sz < ERTS_MIN_PROCESSES
+		    || ERTS_MAX_PROCESSES < proc_tab_sz) {
+		    erts_fprintf(stderr, "bad number of processes %s\n", arg);
+		    erts_usage();
+		}
 	    }
 	    break;
 
 	case 'Q': /* set maximum number of ports */
 	    arg = get_arg(argv[i]+2, argv[i+1], &i);
-	    errno = 0;
-	    port_tab_sz = strtol(arg, NULL, 10);
-	    if (errno != 0
-		|| port_tab_sz < ERTS_MIN_PROCESSES
-		|| ERTS_MAX_PROCESSES < port_tab_sz) {
-		erts_fprintf(stderr, "bad number of ports %s\n", arg);
-		erts_usage();
+	    if (strcmp(arg, "legacy") == 0)
+		legacy_port_tab = 1;
+	    else {
+		errno = 0;
+		port_tab_sz = strtol(arg, NULL, 10);
+		if (errno != 0
+		    || port_tab_sz < ERTS_MIN_PROCESSES
+		    || ERTS_MAX_PROCESSES < port_tab_sz) {
+		    erts_fprintf(stderr, "bad number of ports %s\n", arg);
+		    erts_usage();
+		}
+		port_tab_sz_ignore_files = 1;
 	    }
-	    port_tab_sz_ignore_files = 1;
 	    break;
 
 	case 'S' : /* Was handled in early_init() just read past it */
@@ -1642,8 +1659,10 @@ erl_start(int argc, char **argv)
 
     erl_init(ncpu,
 	     proc_tab_sz,
+	     legacy_proc_tab,
 	     port_tab_sz,
-	     port_tab_sz_ignore_files);
+	     port_tab_sz_ignore_files,
+	     legacy_port_tab);
 
     load_preloaded();
     erts_end_staging_code_ix();
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index f9ef4eac65..5cf18735fd 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -485,7 +485,7 @@ release_process(void *vproc)
 
 /* initialize the scheduler */
 void
-erts_init_process(int ncpu, int proc_tab_size)
+erts_init_process(int ncpu, int proc_tab_size, int legacy_proc_tab)
 {
 
 #ifdef ERTS_SMP
@@ -505,7 +505,8 @@ erts_init_process(int ncpu, int proc_tab_size)
 			 (ErtsPTabElementCommon *) &erts_invalid_process.common,
 			 proc_tab_size,
 			 sizeof(Process),
-			 "process_table");
+			 "process_table",
+			 legacy_proc_tab);
 
     last_reductions = 0;
     last_exact_reductions = 0;
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index 3d3579fa7e..9bf3bf097d 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -1351,7 +1351,7 @@ void erts_schedule_multi_misc_aux_work(int ignore_self,
 erts_aint32_t erts_set_aux_work_timeout(int, erts_aint32_t, int);
 void erts_sched_notify_check_cpu_bind(void);
 Uint erts_active_schedulers(void);
-void erts_init_process(int, int);
+void erts_init_process(int, int, int);
 Eterm erts_process_status(Process *, ErtsProcLocks, Process *, Eterm);
 Uint erts_run_queues_len(Uint *);
 void erts_add_to_runq(Process *);
diff --git a/erts/emulator/beam/erl_ptab.c b/erts/emulator/beam/erl_ptab.c
index 5bbc71c659..8da135b2c8 100644
--- a/erts/emulator/beam/erl_ptab.c
+++ b/erts/emulator/beam/erl_ptab.c
@@ -34,6 +34,8 @@
 
 typedef struct ErtsPTabListBifData_ ErtsPTabListBifData;
 
+#define ERTS_PTAB_NEW_MAX_RESERVE_FAIL 1000
+
 #define ERTS_PTAB_LIST_BIF_TAB_INSPECT_INDICES_PER_RED 25
 #define ERTS_PTAB_LIST_BIF_TAB_CHUNK_SIZE 1000
 #define ERTS_PTAB_LIST_BIF_MIN_START_REDS				\
@@ -415,6 +417,27 @@ last_data_cmp(Uint64 ld1, Uint64 ld2)
 #define ERTS_PTAB_LastData2EtermData(LD) \
     ((Eterm) ((LD) & ~(~((Uint64) 0) << ERTS_PTAB_ID_DATA_SIZE)))
 
+static ERTS_INLINE Uint32
+ix_to_free_id_data_ix(ErtsPTab *ptab, Uint32 ix)
+{
+    Uint32 dix;
+
+    dix = ((ix & ptab->r.o.dix_cl_mask) << ptab->r.o.dix_cl_shift);
+    dix += ((ix >> ptab->r.o.dix_cli_shift) & ptab->r.o.dix_cli_mask);
+    ASSERT(0 <= dix && dix < ptab->r.o.max);
+    return dix;
+}
+
+UWord
+erts_ptab_mem_size(ErtsPTab *ptab)
+{
+    UWord size = ptab->r.o.max*sizeof(erts_smp_atomic_t);
+    if (ptab->r.o.free_id_data)
+	size += ptab->r.o.max*sizeof(Uint32);
+    return size;
+}
+
+
 void
 erts_ptab_init_table(ErtsPTab *ptab,
 		     ErtsAlcType_t atype,
@@ -422,10 +445,11 @@ erts_ptab_init_table(ErtsPTab *ptab,
 		     ErtsPTabElementCommon *invalid_element,
 		     int size,
 		     UWord element_size,
-		     char *name)
+		     char *name,
+		     int legacy)
 {
-    size_t tab_sz;
-    int bits;
+    size_t tab_sz, alloc_sz;
+    Uint32 bits, cl, cli, ix, ix_per_cache_line, tab_cache_lines; 
     char *tab_end;
     erts_smp_atomic_t *tab_entry;
     erts_smp_rwmtx_opt_t rwmtx_opts = ERTS_SMP_RWMTX_OPT_DEFAULT_INITER;
@@ -448,7 +472,10 @@ erts_ptab_init_table(ErtsPTab *ptab,
     ptab->r.o.max = size;
 
     tab_sz = ERTS_ALC_CACHE_LINE_ALIGN_SIZE(size*sizeof(erts_smp_atomic_t));
-    ptab->r.o.tab = erts_alloc_permanent_cache_aligned(atype, tab_sz);
+    alloc_sz = tab_sz;
+    if (!legacy)
+	alloc_sz += ERTS_ALC_CACHE_LINE_ALIGN_SIZE(size*sizeof(Uint32));
+    ptab->r.o.tab = erts_alloc_permanent_cache_aligned(atype, alloc_sz);
     tab_end = ((char *) ptab->r.o.tab) + tab_sz;
     tab_entry = ptab->r.o.tab;
     while (tab_end > ((char *) tab_entry)) {
@@ -456,28 +483,57 @@ erts_ptab_init_table(ErtsPTab *ptab,
 	tab_entry++;
     }
 
-    ptab->r.o.tab_cache_lines = tab_sz/ERTS_CACHE_LINE_SIZE;
-    ptab->r.o.pix_per_cache_line = (ERTS_CACHE_LINE_SIZE
-				    / sizeof(erts_smp_atomic_t));
+    tab_cache_lines = tab_sz/ERTS_CACHE_LINE_SIZE;
+    ix_per_cache_line = (ERTS_CACHE_LINE_SIZE/sizeof(erts_smp_atomic_t));
     ASSERT((ptab->r.o.max & (ptab->r.o.max - 1)) == 0); /* power of 2 */
-    ASSERT((ptab->r.o.pix_per_cache_line
-	    & (ptab->r.o.pix_per_cache_line - 1)) == 0); /* power of 2 */
-    ASSERT((ptab->r.o.tab_cache_lines
-	    & (ptab->r.o.tab_cache_lines - 1)) == 0); /* power of 2 */
-
-    ptab->r.o.pix_mask
-	= (1 << bits) - 1;
-    ptab->r.o.pix_cl_mask
-	= ptab->r.o.tab_cache_lines-1;
-    ptab->r.o.pix_cl_shift
-	= erts_fit_in_bits_int32(ptab->r.o.pix_per_cache_line-1);
-    ptab->r.o.pix_cli_shift
-	= erts_fit_in_bits_int32(ptab->r.o.pix_cl_mask);
-    ptab->r.o.pix_cli_mask
-	= (1 << (bits - ptab->r.o.pix_cli_shift)) - 1;
+    ASSERT((ix_per_cache_line & (ix_per_cache_line - 1)) == 0); /* power of 2 */
+    ASSERT((tab_cache_lines & (tab_cache_lines - 1)) == 0); /* power of 2 */
+
+    ptab->r.o.pix_mask = (1 << bits) - 1;
+    ptab->r.o.pix_cl_mask = tab_cache_lines-1;
+    ptab->r.o.pix_cl_shift = erts_fit_in_bits_int32(ix_per_cache_line-1);
+    ptab->r.o.pix_cli_shift = erts_fit_in_bits_int32(ptab->r.o.pix_cl_mask);
+    ptab->r.o.pix_cli_mask = (1 << (bits - ptab->r.o.pix_cli_shift)) - 1;
 
     ASSERT(ptab->r.o.pix_cl_shift + ptab->r.o.pix_cli_shift == bits);
 
+    if (legacy) {
+	ptab->r.o.free_id_data = NULL;
+	ptab->r.o.dix_cl_mask = 0;
+	ptab->r.o.dix_cl_shift = 0;
+	ptab->r.o.dix_cli_shift = 0;
+	ptab->r.o.dix_cli_mask = 0;
+    }
+    else {
+
+	tab_sz = ERTS_ALC_CACHE_LINE_ALIGN_SIZE(size*sizeof(Uint32));
+	ptab->r.o.free_id_data = (Uint32 *) tab_end;
+
+	tab_cache_lines = tab_sz/ERTS_CACHE_LINE_SIZE;
+	ix_per_cache_line = (ERTS_CACHE_LINE_SIZE/sizeof(Uint32));
+
+	ptab->r.o.dix_cl_mask = tab_cache_lines-1;
+	ptab->r.o.dix_cl_shift = erts_fit_in_bits_int32(ix_per_cache_line-1);
+	ptab->r.o.dix_cli_shift = erts_fit_in_bits_int32(ptab->r.o.dix_cl_mask);
+	ptab->r.o.dix_cli_mask = (1 << (bits - ptab->r.o.dix_cli_shift)) - 1;
+
+	ASSERT((ix_per_cache_line & (ix_per_cache_line - 1)) == 0); /* power of 2 */
+	ASSERT((tab_cache_lines & (tab_cache_lines - 1)) == 0); /* power of 2 */
+
+	ASSERT(ptab->r.o.dix_cl_shift + ptab->r.o.dix_cli_shift == bits);
+
+	ix = 0;
+	for (cl = 0; cl < tab_cache_lines; cl++) {
+	    for (cli = 0; cli < ix_per_cache_line; cli++) {
+		ptab->r.o.free_id_data[ix] = cli*tab_cache_lines+cl;
+		ix++;
+	    }
+	}
+
+	erts_smp_atomic32_init_nob(&ptab->vola.tile.aid_ix, -1);
+	erts_smp_atomic32_init_nob(&ptab->vola.tile.fid_ix, -1);
+
+    }
     ptab->r.o.invalid_element = invalid_element;
     ptab->r.o.invalid_data = erts_ptab_id2data(ptab, invalid_element->id);
     ptab->r.o.release_element = release_element;
@@ -522,9 +578,7 @@ erts_ptab_new_element(ErtsPTab *ptab,
 		      void *init_arg,
 		      void (*init_ptab_el)(void *, Eterm))
 {
-    int pix;
-    Uint64 ld, exp_ld;
-    Eterm data;
+    Uint32 pix, ix, data;
     erts_aint32_t count;
     erts_aint_t invalid = (erts_aint_t) ptab->r.o.invalid_element;
 
@@ -551,62 +605,108 @@ erts_ptab_new_element(ErtsPTab *ptab,
     ptab_el->u.alive.started_interval
 	= erts_smp_current_interval_nob(erts_ptab_interval(ptab));
 
-    ld = last_data_read_acqb(ptab);
+    if (ptab->r.o.free_id_data) {
 
-    /* Reserve slot */
-    while (1) {
-	ld++;
-	pix = erts_ptab_data2pix(ptab, ERTS_PTAB_LastData2EtermData(ld));
-	if (erts_smp_atomic_read_nob(&ptab->r.o.tab[pix]) == ERTS_AINT_NULL) {
-	    erts_aint_t val;
-	    val = erts_smp_atomic_cmpxchg_relb(&ptab->r.o.tab[pix],
-					       invalid,	
-					       ERTS_AINT_NULL);
+	ix = (Uint32) erts_smp_atomic32_inc_read_acqb(&ptab->vola.tile.aid_ix);
+	ix = ix_to_free_id_data_ix(ptab, ix);
+
+	data = ptab->r.o.free_id_data[ix];
+
+	init_ptab_el(init_arg, (Eterm) data);
+
+#ifdef ERTS_SMP
+	erts_smp_atomic32_init_nob(&ptab_el->refc, 1);
+#endif
+
+	pix = erts_ptab_data2pix(ptab, (Eterm) data);
+
+#ifdef DEBUG
+	ASSERT(ERTS_AINT_NULL == erts_smp_atomic_xchg_relb(&ptab->r.o.tab[pix],
+							   (erts_aint_t) ptab_el));
+#else
+	erts_smp_atomic_set_relb(&ptab->r.o.tab[pix], (erts_aint_t) ptab_el);
+#endif
+
+	erts_ptab_runlock(ptab);
 
-	    if (ERTS_AINT_NULL == val)
-		break;
-	}
     }
+    else {
+	int rlocked = ERTS_PTAB_NEW_MAX_RESERVE_FAIL;
+	Uint64 ld, exp_ld;
+	/* Deprecated legacy algorithm... */
+
+    restart:
+
+	ptab_el->u.alive.started_interval
+	    = erts_smp_current_interval_nob(erts_ptab_interval(ptab));
 
-    data = ERTS_PTAB_LastData2EtermData(ld);
+	ld = last_data_read_acqb(ptab);
+
+	/* Reserve slot */
+	while (1) {
+	    ld++;
+	    pix = erts_ptab_data2pix(ptab, ERTS_PTAB_LastData2EtermData(ld));
+	    if (erts_smp_atomic_read_nob(&ptab->r.o.tab[pix])
+		== ERTS_AINT_NULL) {
+		erts_aint_t val;
+		val = erts_smp_atomic_cmpxchg_relb(&ptab->r.o.tab[pix],
+						   invalid,	
+						   ERTS_AINT_NULL);
+
+		if (ERTS_AINT_NULL == val)
+		    break;
+	    }
+	    if (rlocked && --rlocked == 0) {
+		erts_ptab_runlock(ptab);
+		erts_ptab_rwlock(ptab);
+		goto restart;
+	    }
+	}
 
-    if (data == ptab->r.o.invalid_data) {
-	/* Do not use invalid data; fix it... */
-	ld += ptab->r.o.max;
-	ASSERT(pix == erts_ptab_data2pix(ptab,
-					 ERTS_PTAB_LastData2EtermData(ld)));
 	data = ERTS_PTAB_LastData2EtermData(ld);
-	ASSERT(data != ptab->r.o.invalid_data);
-    }
 
-    exp_ld = last_data_read_nob(ptab);
+	if (data == ptab->r.o.invalid_data) {
+	    /* Do not use invalid data; fix it... */
+	    ld += ptab->r.o.max;
+	    ASSERT(pix == erts_ptab_data2pix(ptab,
+					     ERTS_PTAB_LastData2EtermData(ld)));
+	    data = ERTS_PTAB_LastData2EtermData(ld);
+	    ASSERT(data != ptab->r.o.invalid_data);
+	}
 
-    /* Move last data forward */
-    while (1) {
-	Uint64 act_ld;
-	if (last_data_cmp(ld, exp_ld) < 0)
-	    break;
-	act_ld = last_data_cmpxchg_relb(ptab, ld, exp_ld);
-	if (act_ld == exp_ld)
-	    break;
-	exp_ld = act_ld;
-    }
+	exp_ld = last_data_read_nob(ptab);
+
+	/* Move last data forward */
+	while (1) {
+	    Uint64 act_ld;
+	    if (last_data_cmp(ld, exp_ld) < 0)
+		break;
+	    act_ld = last_data_cmpxchg_relb(ptab, ld, exp_ld);
+	    if (act_ld == exp_ld)
+		break;
+	    exp_ld = act_ld;
+	}
 
-    init_ptab_el(init_arg, data);
+	init_ptab_el(init_arg, data);
 
 #ifdef ERTS_SMP
-    erts_smp_atomic32_init_nob(&ptab_el->refc, 1);
+	erts_smp_atomic32_init_nob(&ptab_el->refc, 1);
 #endif
 
-    /* Move into slot reserved */
+	/* Move into slot reserved */
 #ifdef DEBUG
-    ASSERT(invalid == erts_smp_atomic_xchg_relb(&ptab->r.o.tab[pix],
+	ASSERT(invalid == erts_smp_atomic_xchg_relb(&ptab->r.o.tab[pix],
 						(erts_aint_t) ptab_el));
 #else
-    erts_smp_atomic_set_relb(&ptab->r.o.tab[pix], (erts_aint_t) ptab_el);
+	erts_smp_atomic_set_relb(&ptab->r.o.tab[pix], (erts_aint_t) ptab_el);
 #endif
 
-    erts_ptab_runlock(ptab);
+	if (rlocked)
+	    erts_ptab_runlock(ptab);
+	else
+	    erts_ptab_rwunlock(ptab);
+
+    }
 
     return 1;
 }
@@ -647,7 +747,9 @@ erts_ptab_delete_element(ErtsPTab *ptab,
 			 ErtsPTabElementCommon *ptab_el)
 {
     int maybe_save;
-    int pix = erts_ptab_id2pix(ptab, ptab_el->id);
+    Uint32 pix, ix, data;
+
+    pix = erts_ptab_id2pix(ptab, ptab_el->id);
 
     ASSERT(erts_get_scheduler_id()); /* *Need* to be a scheduler */
 
@@ -660,6 +762,26 @@ erts_ptab_delete_element(ErtsPTab *ptab,
 
     erts_smp_atomic_set_relb(&ptab->r.o.tab[pix], ERTS_AINT_NULL);
 
+    if (ptab->r.o.free_id_data) {
+
+	/* Next data for this slot... */
+	data = (Uint32) erts_ptab_id2data(ptab, ptab_el->id);
+	data += ptab->r.o.max;
+	data &= ~(~((Uint32) 0) << ERTS_PTAB_ID_DATA_SIZE);
+	if (data == ptab->r.o.invalid_data) { /* make sure not invalid */
+	    data += ptab->r.o.max;
+	    data &= ~(~((Uint32) 0) << ERTS_PTAB_ID_DATA_SIZE);
+	}
+
+	ASSERT(data != ptab->r.o.invalid_data);
+	ASSERT(pix == erts_ptab_data2pix(ptab, data));
+
+	ix = (Uint32) erts_smp_atomic32_inc_read_relb(&ptab->vola.tile.fid_ix);
+	ix = ix_to_free_id_data_ix(ptab, ix);
+
+	ptab->r.o.free_id_data[ix] = data;
+    }
+
     ASSERT(erts_smp_atomic32_read_nob(&ptab->vola.tile.count) > 0);
     erts_smp_atomic32_dec_relb(&ptab->vola.tile.count);
 
@@ -1280,42 +1402,86 @@ erts_ptab_test_next_id(ErtsPTab *ptab, int set, Uint next)
 
     erts_ptab_rwlock(ptab);
 
-    if (!set)
-	ld = last_data_read_nob(ptab);
-    else {
+    if (ptab->r.o.free_id_data) {
+	Uint32 aid_ix, dix;
 
-	ld = (Uint64) next;
-	data = ERTS_PTAB_LastData2EtermData(ld);
-	if (ptab->r.o.invalid_data == data) {
-	    ld += ptab->r.o.max;
-	    ASSERT(erts_ptab_data2pix(ptab, data)
-		   == erts_ptab_data2pix(ptab,
-					 ERTS_PTAB_LastData2EtermData(ld)));
+	if (set) {
+	    Uint32 max_ix, ser, num, start;
+	    max_ix = ptab->r.o.max - 1;
+	    ser = next & ~max_ix;
+	    start = num = next & max_ix;
+	
+	    aid_ix = (Uint32) erts_smp_atomic32_read_nob(&ptab->vola.tile.aid_ix) + 1;
+
+	    do {
+		Uint32 pix = erts_ptab_data2pix(ptab, num);
+		if (ERTS_AINT_NULL == erts_ptab_pix2intptr_nob(ptab, pix)) {
+		    dix = ix_to_free_id_data_ix(ptab, aid_ix);
+		    ptab->r.o.free_id_data[dix] = ser + num;
+		    ASSERT(pix == erts_ptab_data2pix(ptab, ser+num));
+		    if (aid_ix == max_ix)
+			aid_ix = 0;
+		    else
+			aid_ix++;
+		}
+		if (num == max_ix)
+		    num = 0;
+		else
+		    num++;
+	    } while (num != start);
+
+#ifdef DEBUG
+	    if (aid_ix == 0)
+		aid_ix = max_ix;
+	    else
+		aid_ix--;
+	    ASSERT((aid_ix & max_ix) == (((Uint32) erts_atomic32_read_nob(&ptab->vola.tile.fid_ix)) & max_ix));
+#endif
 	}
-	last_data_set_relb(ptab, ld);
+
+	aid_ix = (Uint32) erts_smp_atomic32_read_nob(&ptab->vola.tile.aid_ix) + 1;
+	dix = ix_to_free_id_data_ix(ptab, aid_ix);
+	res = (Sint) ptab->r.o.free_id_data[dix];
     }
+    else {
+	/* Deprecated legacy algorithm... */
+	if (!set)
+	    ld = last_data_read_nob(ptab);
+	else {
 
-    while (1) {
-	int pix;
-	ld++;
-	pix = (int) (ld % ptab->r.o.max);
-	if (first_pix < 0)
-	    first_pix = pix;
-	else if (pix == first_pix) {
-	    res = -1;
-	    break;
-	}
-	if (ERTS_AINT_NULL == erts_ptab_pix2intptr_nob(ptab, pix)) {
+	    ld = (Uint64) next;
 	    data = ERTS_PTAB_LastData2EtermData(ld);
 	    if (ptab->r.o.invalid_data == data) {
 		ld += ptab->r.o.max;
 		ASSERT(erts_ptab_data2pix(ptab, data)
 		       == erts_ptab_data2pix(ptab,
 					     ERTS_PTAB_LastData2EtermData(ld)));
+	    }
+	    last_data_set_relb(ptab, ld);
+	}
+
+	while (1) {
+	    int pix;
+	    ld++;
+	    pix = (int) (ld % ptab->r.o.max);
+	    if (first_pix < 0)
+		first_pix = pix;
+	    else if (pix == first_pix) {
+		res = -1;
+		break;
+	    }
+	    if (ERTS_AINT_NULL == erts_ptab_pix2intptr_nob(ptab, pix)) {
 		data = ERTS_PTAB_LastData2EtermData(ld);
+		if (ptab->r.o.invalid_data == data) {
+		    ld += ptab->r.o.max;
+		    ASSERT(erts_ptab_data2pix(ptab, data)
+			   == erts_ptab_data2pix(ptab,
+						 ERTS_PTAB_LastData2EtermData(ld)));
+		    data = ERTS_PTAB_LastData2EtermData(ld);
+		}
+		res = data;
+		break;
 	    }
-	    res = data;
-	    break;
 	}
     }
 
diff --git a/erts/emulator/beam/erl_ptab.h b/erts/emulator/beam/erl_ptab.h
index 7fa1251900..84ff7d0de4 100644
--- a/erts/emulator/beam/erl_ptab.h
+++ b/erts/emulator/beam/erl_ptab.h
@@ -94,18 +94,23 @@ typedef struct {
     erts_smp_atomic_t last_data;
 #endif
     erts_smp_atomic32_t count;
+    erts_smp_atomic32_t aid_ix;
+    erts_smp_atomic32_t fid_ix;
 } ErtsPTabVolatileData;
 
 typedef struct {
     erts_smp_atomic_t *tab;
+    Uint32 *free_id_data;
     Uint32 max;
-    Uint32 tab_cache_lines;
-    Uint32 pix_per_cache_line;
     Uint32 pix_mask;
     Uint32 pix_cl_mask;
     Uint32 pix_cl_shift;
     Uint32 pix_cli_mask;
     Uint32 pix_cli_shift;
+    Uint32 dix_cl_mask;
+    Uint32 dix_cl_shift;
+    Uint32 dix_cli_mask;
+    Uint32 dix_cli_shift;
     ErtsPTabElementCommon *invalid_element;
     Eterm invalid_data;
     void (*release_element)(void *);
@@ -179,7 +184,8 @@ void erts_ptab_init_table(ErtsPTab *ptab,
 			  ErtsPTabElementCommon *invalid_element,
 			  int size,
 			  UWord element_size,
-			  char *name);
+			  char *name,
+			  int legacy);
 int erts_ptab_new_element(ErtsPTab *ptab,
 			  ErtsPTabElementCommon *ptab_el,
 			  void *init_arg,
@@ -187,6 +193,7 @@ int erts_ptab_new_element(ErtsPTab *ptab,
 void erts_ptab_delete_element(ErtsPTab *ptab,
 			      ErtsPTabElementCommon *ptab_el);
 int erts_ptab_initialized(ErtsPTab *ptab);
+UWord erts_ptab_mem_size(ErtsPTab *ptab);
 
 ERTS_GLB_INLINE erts_interval_t *erts_ptab_interval(ErtsPTab *ptab);
 ERTS_GLB_INLINE int erts_ptab_max(ErtsPTab *ptab);
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index 012c1c7e6a..c900cdca39 100755
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -724,7 +724,7 @@ int erts_add_driver_entry(ErlDrvEntry *drv, DE_Handle *handle, int driver_list_l
 void erts_destroy_driver(erts_driver_t *drv);
 int erts_save_suspend_process_on_port(Port*, Process*);
 Port *erts_open_driver(erts_driver_t*, Eterm, char*, SysDriverOpts*, int *, int *);
-void erts_init_io(int, int);
+void erts_init_io(int, int, int);
 void erts_raw_port_command(Port*, byte*, Uint);
 void driver_report_exit(ErlDrvPort, int);
 LineBuf* allocate_linebuf(int);
diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c
index 43918a7141..e89dd89f6a 100644
--- a/erts/emulator/beam/io.c
+++ b/erts/emulator/beam/io.c
@@ -2648,7 +2648,8 @@ erts_port_link(Process *c_p, Port *prt, Eterm to, Eterm *refp)
 }
 
 void erts_init_io(int port_tab_size,
-		  int port_tab_size_ignore_files)
+		  int port_tab_size_ignore_files,
+		  int legacy_port_tab)
 {
     ErlDrvEntry** dp;
     UWord common_element_size;
@@ -2691,7 +2692,8 @@ void erts_init_io(int port_tab_size,
 			 (ErtsPTabElementCommon *) &erts_invalid_port.common,
 			 port_tab_size,
 			 common_element_size, /* Doesn't need to be excact */
-			 "port_table");
+			 "port_table",
+			 legacy_port_tab);
 
     erts_smp_atomic_init_nob(&erts_bytes_out, 0);
     erts_smp_atomic_init_nob(&erts_bytes_in, 0);
diff --git a/erts/emulator/test/process_SUITE.erl b/erts/emulator/test/process_SUITE.erl
index 863cd2d654..72f3e8fe85 100644
--- a/erts/emulator/test/process_SUITE.erl
+++ b/erts/emulator/test/process_SUITE.erl
@@ -1494,6 +1494,7 @@ processes_bif_cleaner() ->
 
 spawn_initial_hangarounds(Cleaner) ->
     TabSz = erlang:system_info(process_limit),
+    erts_debug:set_internal_state(next_pid,TabSz),
     spawn_initial_hangarounds(Cleaner,
 			      TabSz,
 			      TabSz*2,
@@ -1538,14 +1539,21 @@ hangaround(Cleaner, Type) ->
 spawn_initial_hangarounds(_Cleaner, NP, Max, Len, HAs) when NP > Max ->
     {Len, HAs};
 spawn_initial_hangarounds(Cleaner, NP, Max, Len, HAs) ->
-    erts_debug:set_internal_state(next_pid,NP),
+    Skip = 30,
     HA1 = spawn_opt(?MODULE, hangaround, [Cleaner, initial_hangaround],
 		    [{priority, low}]),
     HA2 = spawn_opt(?MODULE, hangaround, [Cleaner, initial_hangaround],
 		    [{priority, normal}]),
     HA3 = spawn_opt(?MODULE, hangaround, [Cleaner, initial_hangaround],
 		    [{priority, high}]),
-    spawn_initial_hangarounds(Cleaner, NP+30, Max, Len+3, [HA1,HA2,HA3|HAs]).
+    spawn_drop(Skip),
+    spawn_initial_hangarounds(Cleaner, NP+Skip, Max, Len+3, [HA1,HA2,HA3|HAs]).
+
+spawn_drop(N) when N =< 0 ->
+    ok;
+spawn_drop(N) ->
+    spawn(fun () -> ok end),
+    spawn_drop(N-1).
 
 do_processes(WantReds) ->
     erts_debug:set_internal_state(reds_left, WantReds),