diff mbox

[dpdk-dev,v4,3/7] member: implement vBF mode

Message ID 1506534034-39433-4-git-send-email-yipeng1.wang@intel.com (mailing list archive)
State Superseded, archived
Headers show

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK

Commit Message

Wang, Yipeng1 Sept. 27, 2017, 5:40 p.m. UTC
Bloom Filter (BF) [1] is a well-known space-efficient
probabilistic data structure that answers set membership queries.
Vector of Bloom Filters (vBF) is an extension to traditional BF
that supports multi-set membership testing. Traditional BF will
return found or not-found for each key. vBF will also return
which set the key belongs to if it is found.

Since each set requires a BF, vBF should be used when set count
is small. vBF's false positive rate could be set appropriately so
that its memory requirement and lookup speed is better in certain
cases comparing to HT based set-summary.

This patch adds the vBF implementation.

[1]B H Bloom, “Space/Time Trade-offs in Hash Coding with Allowable
Errors,” Communications of the ACM, 1970.

Signed-off-by: Yipeng Wang <yipeng1.wang@intel.com>
---
 lib/librte_member/Makefile         |   2 +-
 lib/librte_member/rte_member_vbf.c | 349 +++++++++++++++++++++++++++++++++++++
 lib/librte_member/rte_member_vbf.h |  79 +++++++++
 3 files changed, 429 insertions(+), 1 deletion(-)
 create mode 100644 lib/librte_member/rte_member_vbf.c
 create mode 100644 lib/librte_member/rte_member_vbf.h

Comments

Pablo de Lara Oct. 2, 2017, 3:44 p.m. UTC | #1
> -----Original Message-----

> From: Wang, Yipeng1

> Sent: Wednesday, September 27, 2017 6:41 PM

> To: dev@dpdk.org

> Cc: thomas@monjalon.net; Tai, Charlie <charlie.tai@intel.com>; Gobriel,

> Sameh <sameh.gobriel@intel.com>; De Lara Guarch, Pablo

> <pablo.de.lara.guarch@intel.com>; Mcnamara, John

> <john.mcnamara@intel.com>; Wang, Yipeng1 <yipeng1.wang@intel.com>

> Subject: [PATCH v4 3/7] member: implement vBF mode

> 

> Bloom Filter (BF) [1] is a well-known space-efficient probabilistic data

> structure that answers set membership queries.

> Vector of Bloom Filters (vBF) is an extension to traditional BF that supports

> multi-set membership testing. Traditional BF will return found or not-found

> for each key. vBF will also return which set the key belongs to if it is found.

> 

> Since each set requires a BF, vBF should be used when set count is small.

> vBF's false positive rate could be set appropriately so that its memory

> requirement and lookup speed is better in certain cases comparing to HT

> based set-summary.

> 

> This patch adds the vBF implementation.

> 

> [1]B H Bloom, “Space/Time Trade-offs in Hash Coding with Allowable

> Errors,” Communications of the ACM, 1970.

> 

> Signed-off-by: Yipeng Wang <yipeng1.wang@intel.com>


...

> diff --git a/lib/librte_member/rte_member_vbf.c

> b/lib/librte_member/rte_member_vbf.c


...

> +int

> +rte_member_create_vbf(struct rte_member_setsum *ss,

> +		const struct rte_member_parameters *params) {

> +

> +	if (params->num_set > 32 || !rte_is_power_of_2(params-

> >num_set) ||


Magic number. Define a macro instead.

> +			params->num_keys == 0 ||

> +			params->false_positive_rate == 0 ||

> +			params->false_positive_rate > 1) {

> +		rte_errno = EINVAL;

> +		RTE_MEMBER_LOG(ERR, "vBF create with invalid

> parameters\n");

> +		return -EINVAL;


...

> +

> +	/*

> +	 * reduce hash function count, until we approach the user specified

> +	 * false-positive rate. otherwise it is too conservative


Watch out for capital letters at the start of the comment and after a full stop.

> +	 */

> +	int tmp_num_hash = ss->num_hashes;

> +

> +	while (tmp_num_hash > 1) {

> +		float tmp_fp = new_fp;

> +

> +		tmp_num_hash--;

> +		new_fp = pow((1 - pow((1 - 1.0 / ss->bits),

> num_keys_per_bf *

> +					tmp_num_hash)), tmp_num_hash);

> +		new_fp = 1 - pow((1 - new_fp), ss->num_set);

> +

> +		if (new_fp > params->false_positive_rate) {

> +			new_fp = tmp_fp;

> +			tmp_num_hash++;

> +			break;

> +		}

> +	}

> +

> +	ss->num_hashes = tmp_num_hash;

> +

> +	RTE_MEMBER_LOG(DEBUG, "vector bloom filter created, "

> +		"each bloom filter expects %u keys, needs %u bits, %u

> hashes, "

> +		"with false positive rate set as %.5f, "

> +		"The new calculated vBF false positive rate is %.5f\n",

> +		num_keys_per_bf, ss->bits, ss->num_hashes, x, new_fp);


Use a more descriptive variable name for "x".

> +

> +	ss->table = rte_zmalloc_socket(NULL, ss->num_set * (ss->bits >> 3),

> +					RTE_CACHE_LINE_SIZE, ss-

> >socket_id);

> +

> +	/*

> +	 * To avoid multiplication and division:

> +	 * mul_shift is used for multiplication shift during bit test

> +	 * div_shift is used for division shift, to be divided by number of bits

> +	 * represented by a uint32_t variable

> +	 */

> +	ss->mul_shift = __builtin_ctzl(ss->num_set);

> +	ss->div_shift = __builtin_ctzl(32 >> ss->mul_shift);

> +

> +	if (ss->table == NULL)

> +		return -ENOMEM;


I would move this check just after the malloc call.

> +

> +	return 0;

> +}

> +

> +static inline uint32_t

> +test_bit(uint32_t bit_loc, const struct rte_member_setsum *ss) {

> +	uint32_t *vbf = ss->table;

> +	uint32_t n = ss->num_set;

> +	uint32_t div_shift = ss->div_shift;

> +	uint32_t mul_shift = ss->mul_shift;

> +	/*

> +	 * a is how many bits in one BF are represented by one 32bit

> +	 * variable.

> +	 */

> +	uint32_t a = 32 >> mul_shift;

> +	/*

> +	 * x>>b is the divide, x & (a-1) is the mod, & (1<<n-1) to mask out

> bits

> +	 * we do not need

> +	 */

> +	return (vbf[bit_loc>>div_shift] >> ((bit_loc & (a - 1)) << mul_shift))


Add spaces around ">>".

> &

> +							((1ULL << n) - 1);

> +}

> +

> +static inline void

> +set_bit(uint32_t bit_loc, const struct rte_member_setsum *ss, int32_t

> +set) {

> +	uint32_t *vbf = ss->table;

> +	uint32_t div_shift = ss->div_shift;

> +	uint32_t mul_shift = ss->mul_shift;

> +	uint32_t a = 32 >> mul_shift;

> +

> +	vbf[bit_loc>>div_shift] |= 1U << (((bit_loc & (a - 1)) << mul_shift) +

> +								set - 1);


Same as above.

> +}

> +

> +int

> +rte_member_lookup_vbf(const struct rte_member_setsum *ss, const

> void *key,

> +		member_set_t *set_id)

> +{

> +	uint32_t j;

> +	uint32_t h1 = MEMBER_HASH_FUNC(key, ss->key_len, ss-

> >prim_hash_seed);

> +	uint32_t h2 = MEMBER_HASH_FUNC(&h1, sizeof(uint32_t),

> +						ss->sec_hash_seed);

> +	uint32_t mask = ~0;

> +	uint32_t bit_loc;

> +

> +	for (j = 0; j < ss->num_hashes; j++) {

> +		bit_loc = (h1 + j * h2) & ss->bit_mask;

> +		mask &= test_bit(bit_loc, ss);

> +	}

> +

> +	if (mask) {

> +		*set_id = __builtin_ctzl(mask) + 1;

> +		return 1;


Wouldn't it be better to return 0 when there is a hit and -ENOENT when there is not?
> +	}

> +

> +	*set_id = RTE_MEMBER_NO_MATCH;

> +	return 0;

> +}

> +

> +uint32_t

> +rte_member_lookup_bulk_vbf(const struct rte_member_setsum *ss,

> +		const void **keys, uint32_t num_keys, member_set_t

> *set_ids) {

> +	uint32_t i, k;

> +	uint32_t ret = 0;


Change variable name to "nr_matches/hits" or similar.
Same in the next functions.
Wang, Yipeng1 Oct. 3, 2017, 1:24 a.m. UTC | #2
Please see one comment inlined:

> -----Original Message-----

> From: De Lara Guarch, Pablo

> Sent: Monday, October 2, 2017 8:44 AM

> To: Wang, Yipeng1 <yipeng1.wang@intel.com>; dev@dpdk.org

> Cc: thomas@monjalon.net; Tai, Charlie <charlie.tai@intel.com>; Gobriel,

> Sameh <sameh.gobriel@intel.com>; Mcnamara, John

> <john.mcnamara@intel.com>

> Subject: RE: [PATCH v4 3/7] member: implement vBF mode

> 

> 

> 

> > -----Original Message-----

> > From: Wang, Yipeng1

> > Sent: Wednesday, September 27, 2017 6:41 PM

> > To: dev@dpdk.org

> > Cc: thomas@monjalon.net; Tai, Charlie <charlie.tai@intel.com>; Gobriel,

> > Sameh <sameh.gobriel@intel.com>; De Lara Guarch, Pablo

> > <pablo.de.lara.guarch@intel.com>; Mcnamara, John

> > <john.mcnamara@intel.com>; Wang, Yipeng1 <yipeng1.wang@intel.com>

> > Subject: [PATCH v4 3/7] member: implement vBF mode

> >

> > Bloom Filter (BF) [1] is a well-known space-efficient probabilistic data

> > structure that answers set membership queries.

> > Vector of Bloom Filters (vBF) is an extension to traditional BF that supports

> > multi-set membership testing. Traditional BF will return found or not-found

> > for each key. vBF will also return which set the key belongs to if it is found.

> >

> > Since each set requires a BF, vBF should be used when set count is small.

> > vBF's false positive rate could be set appropriately so that its memory

> > requirement and lookup speed is better in certain cases comparing to HT

> > based set-summary.

> >

> > This patch adds the vBF implementation.

> >

> > [1]B H Bloom, “Space/Time Trade-offs in Hash Coding with Allowable

> > Errors,” Communications of the ACM, 1970.

> >

> > Signed-off-by: Yipeng Wang <yipeng1.wang@intel.com>

> 

> ...

> 

> > diff --git a/lib/librte_member/rte_member_vbf.c

> > b/lib/librte_member/rte_member_vbf.c

> 

> ...

> 

> > +int

> > +rte_member_create_vbf(struct rte_member_setsum *ss,

> > +		const struct rte_member_parameters *params) {

> > +

> > +	if (params->num_set > 32 || !rte_is_power_of_2(params-

> > >num_set) ||

> 

> Magic number. Define a macro instead.

> 

> > +			params->num_keys == 0 ||

> > +			params->false_positive_rate == 0 ||

> > +			params->false_positive_rate > 1) {

> > +		rte_errno = EINVAL;

> > +		RTE_MEMBER_LOG(ERR, "vBF create with invalid

> > parameters\n");

> > +		return -EINVAL;

> 

> ...

> 

> > +

> > +	/*

> > +	 * reduce hash function count, until we approach the user specified

> > +	 * false-positive rate. otherwise it is too conservative

> 

> Watch out for capital letters at the start of the comment and after a full stop.

> 

> > +	 */

> > +	int tmp_num_hash = ss->num_hashes;

> > +

> > +	while (tmp_num_hash > 1) {

> > +		float tmp_fp = new_fp;

> > +

> > +		tmp_num_hash--;

> > +		new_fp = pow((1 - pow((1 - 1.0 / ss->bits),

> > num_keys_per_bf *

> > +					tmp_num_hash)), tmp_num_hash);

> > +		new_fp = 1 - pow((1 - new_fp), ss->num_set);

> > +

> > +		if (new_fp > params->false_positive_rate) {

> > +			new_fp = tmp_fp;

> > +			tmp_num_hash++;

> > +			break;

> > +		}

> > +	}

> > +

> > +	ss->num_hashes = tmp_num_hash;

> > +

> > +	RTE_MEMBER_LOG(DEBUG, "vector bloom filter created, "

> > +		"each bloom filter expects %u keys, needs %u bits, %u

> > hashes, "

> > +		"with false positive rate set as %.5f, "

> > +		"The new calculated vBF false positive rate is %.5f\n",

> > +		num_keys_per_bf, ss->bits, ss->num_hashes, x, new_fp);

> 

> Use a more descriptive variable name for "x".

> 

> > +

> > +	ss->table = rte_zmalloc_socket(NULL, ss->num_set * (ss->bits >> 3),

> > +					RTE_CACHE_LINE_SIZE, ss-

> > >socket_id);

> > +

> > +	/*

> > +	 * To avoid multiplication and division:

> > +	 * mul_shift is used for multiplication shift during bit test

> > +	 * div_shift is used for division shift, to be divided by number of bits

> > +	 * represented by a uint32_t variable

> > +	 */

> > +	ss->mul_shift = __builtin_ctzl(ss->num_set);

> > +	ss->div_shift = __builtin_ctzl(32 >> ss->mul_shift);

> > +

> > +	if (ss->table == NULL)

> > +		return -ENOMEM;

> 

> I would move this check just after the malloc call.

> 

> > +

> > +	return 0;

> > +}

> > +

> > +static inline uint32_t

> > +test_bit(uint32_t bit_loc, const struct rte_member_setsum *ss) {

> > +	uint32_t *vbf = ss->table;

> > +	uint32_t n = ss->num_set;

> > +	uint32_t div_shift = ss->div_shift;

> > +	uint32_t mul_shift = ss->mul_shift;

> > +	/*

> > +	 * a is how many bits in one BF are represented by one 32bit

> > +	 * variable.

> > +	 */

> > +	uint32_t a = 32 >> mul_shift;

> > +	/*

> > +	 * x>>b is the divide, x & (a-1) is the mod, & (1<<n-1) to mask out

> > bits

> > +	 * we do not need

> > +	 */

> > +	return (vbf[bit_loc>>div_shift] >> ((bit_loc & (a - 1)) << mul_shift))

> 

> Add spaces around ">>".

> 

> > &

> > +							((1ULL << n) - 1);

> > +}

> > +

> > +static inline void

> > +set_bit(uint32_t bit_loc, const struct rte_member_setsum *ss, int32_t

> > +set) {

> > +	uint32_t *vbf = ss->table;

> > +	uint32_t div_shift = ss->div_shift;

> > +	uint32_t mul_shift = ss->mul_shift;

> > +	uint32_t a = 32 >> mul_shift;

> > +

> > +	vbf[bit_loc>>div_shift] |= 1U << (((bit_loc & (a - 1)) << mul_shift) +

> > +								set - 1);

> 

> Same as above.

> 

> > +}

> > +

> > +int

> > +rte_member_lookup_vbf(const struct rte_member_setsum *ss, const

> > void *key,

> > +		member_set_t *set_id)

> > +{

> > +	uint32_t j;

> > +	uint32_t h1 = MEMBER_HASH_FUNC(key, ss->key_len, ss-

> > >prim_hash_seed);

> > +	uint32_t h2 = MEMBER_HASH_FUNC(&h1, sizeof(uint32_t),

> > +						ss->sec_hash_seed);

> > +	uint32_t mask = ~0;

> > +	uint32_t bit_loc;

> > +

> > +	for (j = 0; j < ss->num_hashes; j++) {

> > +		bit_loc = (h1 + j * h2) & ss->bit_mask;

> > +		mask &= test_bit(bit_loc, ss);

> > +	}

> > +

> > +	if (mask) {

> > +		*set_id = __builtin_ctzl(mask) + 1;

> > +		return 1;

> 

> Wouldn't it be better to return 0 when there is a hit and -ENOENT when

> there is not?

[Wang, Yipeng] 
I tried to follow the convention of other lookup functions that the return value
Is kind of the number of matches that found. So 1 means found 1 and 0 means not
Found. I defined the public API rte_member_lookup to also returns 1 for found.

> > +	}

> > +

> > +	*set_id = RTE_MEMBER_NO_MATCH;

> > +	return 0;

> > +}

> > +

> > +uint32_t

> > +rte_member_lookup_bulk_vbf(const struct rte_member_setsum *ss,

> > +		const void **keys, uint32_t num_keys, member_set_t

> > *set_ids) {

> > +	uint32_t i, k;

> > +	uint32_t ret = 0;

> 

> Change variable name to "nr_matches/hits" or similar.

> Same in the next functions.
diff mbox

Patch

diff --git a/lib/librte_member/Makefile b/lib/librte_member/Makefile
index ad26548..50275ed 100644
--- a/lib/librte_member/Makefile
+++ b/lib/librte_member/Makefile
@@ -42,7 +42,7 @@  EXPORT_MAP := rte_member_version.map
 LIBABIVER := 1
 
 # all source are stored in SRCS-y
-SRCS-$(CONFIG_RTE_LIBRTE_MEMBER) +=  rte_member.c rte_member_ht.c
+SRCS-$(CONFIG_RTE_LIBRTE_MEMBER) +=  rte_member.c rte_member_ht.c rte_member_vbf.c
 # install includes
 SYMLINK-$(CONFIG_RTE_LIBRTE_MEMBER)-include := rte_member.h
 
diff --git a/lib/librte_member/rte_member_vbf.c b/lib/librte_member/rte_member_vbf.c
new file mode 100644
index 0000000..cbfa18e
--- /dev/null
+++ b/lib/librte_member/rte_member_vbf.c
@@ -0,0 +1,349 @@ 
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2017 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * 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.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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 COPYRIGHT
+ *   OWNER 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.
+ */
+
+#include <math.h>
+#include <string.h>
+
+#include <rte_malloc.h>
+#include <rte_memory.h>
+#include <rte_errno.h>
+#include <rte_log.h>
+
+#include "rte_member.h"
+#include "rte_member_vbf.h"
+
+/*
+ * vBF currently implemented as a big array.
+ * The BFs have a vertical layout. Bits in same location of all bfs will stay
+ * in the same cache line.
+ * For example, if we have 32 bloom filters, we use a uint32_t array to
+ * represent all of them. array[0] represent the first location of all the
+ * bloom filters, array[1] represents the second location of all the
+ * bloom filters, etc. The advantage of this layout is to minimize the average
+ * number of memory accesses to test all bloom filters.
+ *
+ * Currently the implementation supports vBF containing 1,2,4,8,16,32 BFs.
+ */
+int
+rte_member_create_vbf(struct rte_member_setsum *ss,
+		const struct rte_member_parameters *params)
+{
+
+	if (params->num_set > 32 || !rte_is_power_of_2(params->num_set) ||
+			params->num_keys == 0 ||
+			params->false_positive_rate == 0 ||
+			params->false_positive_rate > 1) {
+		rte_errno = EINVAL;
+		RTE_MEMBER_LOG(ERR, "vBF create with invalid parameters\n");
+		return -EINVAL;
+	}
+
+	/* We assume expected keys evenly distribute to all BFs */
+	uint32_t num_keys_per_bf = 1 + (params->num_keys - 1) / ss->num_set;
+
+	/*
+	 * Note that the false positive rate is for all BFs in the vBF
+	 * such that the single BF's false positive rate needs to be
+	 * calculated.
+	 * Assume each BF's False positive rate is x. The total false positive
+	 * rate is fp = 1-(1-x)^n.
+	 * => x = 1 - (1-fp)^(1/n)
+	 */
+
+	float x = 1 - pow((1 - params->false_positive_rate), 1.0 / ss->num_set);
+
+	if (x == 0) {
+		rte_errno = EINVAL;
+		RTE_MEMBER_LOG(ERR, "BF false positive rate is too small\n");
+		return -EINVAL;
+	}
+
+	uint32_t bits = ceil((num_keys_per_bf *
+				log(x)) / log(1.0 / (pow(2.0, log(2.0)))));
+
+	/* We round to power of 2 for performance during lookup */
+	ss->bits = rte_align32pow2(bits);
+
+	ss->num_hashes = (uint32_t)(log(2.0) * bits / num_keys_per_bf);
+	ss->bit_mask = ss->bits - 1;
+
+
+	/*
+	 * Since we round the bits to power of 2, the final false positive
+	 * rate will probably not be same as the user specified. We log the
+	 * new value as debug message.
+	 */
+	float new_fp = pow((1 - pow((1 - 1.0 / ss->bits), num_keys_per_bf *
+					ss->num_hashes)), ss->num_hashes);
+	new_fp = 1 - pow((1 - new_fp), ss->num_set);
+
+	/*
+	 * reduce hash function count, until we approach the user specified
+	 * false-positive rate. otherwise it is too conservative
+	 */
+	int tmp_num_hash = ss->num_hashes;
+
+	while (tmp_num_hash > 1) {
+		float tmp_fp = new_fp;
+
+		tmp_num_hash--;
+		new_fp = pow((1 - pow((1 - 1.0 / ss->bits), num_keys_per_bf *
+					tmp_num_hash)), tmp_num_hash);
+		new_fp = 1 - pow((1 - new_fp), ss->num_set);
+
+		if (new_fp > params->false_positive_rate) {
+			new_fp = tmp_fp;
+			tmp_num_hash++;
+			break;
+		}
+	}
+
+	ss->num_hashes = tmp_num_hash;
+
+	RTE_MEMBER_LOG(DEBUG, "vector bloom filter created, "
+		"each bloom filter expects %u keys, needs %u bits, %u hashes, "
+		"with false positive rate set as %.5f, "
+		"The new calculated vBF false positive rate is %.5f\n",
+		num_keys_per_bf, ss->bits, ss->num_hashes, x, new_fp);
+
+	ss->table = rte_zmalloc_socket(NULL, ss->num_set * (ss->bits >> 3),
+					RTE_CACHE_LINE_SIZE, ss->socket_id);
+
+	/*
+	 * To avoid multiplication and division:
+	 * mul_shift is used for multiplication shift during bit test
+	 * div_shift is used for division shift, to be divided by number of bits
+	 * represented by a uint32_t variable
+	 */
+	ss->mul_shift = __builtin_ctzl(ss->num_set);
+	ss->div_shift = __builtin_ctzl(32 >> ss->mul_shift);
+
+	if (ss->table == NULL)
+		return -ENOMEM;
+
+	return 0;
+}
+
+static inline uint32_t
+test_bit(uint32_t bit_loc, const struct rte_member_setsum *ss)
+{
+	uint32_t *vbf = ss->table;
+	uint32_t n = ss->num_set;
+	uint32_t div_shift = ss->div_shift;
+	uint32_t mul_shift = ss->mul_shift;
+	/*
+	 * a is how many bits in one BF are represented by one 32bit
+	 * variable.
+	 */
+	uint32_t a = 32 >> mul_shift;
+	/*
+	 * x>>b is the divide, x & (a-1) is the mod, & (1<<n-1) to mask out bits
+	 * we do not need
+	 */
+	return (vbf[bit_loc>>div_shift] >> ((bit_loc & (a - 1)) << mul_shift)) &
+							((1ULL << n) - 1);
+}
+
+static inline void
+set_bit(uint32_t bit_loc, const struct rte_member_setsum *ss, int32_t set)
+{
+	uint32_t *vbf = ss->table;
+	uint32_t div_shift = ss->div_shift;
+	uint32_t mul_shift = ss->mul_shift;
+	uint32_t a = 32 >> mul_shift;
+
+	vbf[bit_loc>>div_shift] |= 1U << (((bit_loc & (a - 1)) << mul_shift) +
+								set - 1);
+}
+
+int
+rte_member_lookup_vbf(const struct rte_member_setsum *ss, const void *key,
+		member_set_t *set_id)
+{
+	uint32_t j;
+	uint32_t h1 = MEMBER_HASH_FUNC(key, ss->key_len, ss->prim_hash_seed);
+	uint32_t h2 = MEMBER_HASH_FUNC(&h1, sizeof(uint32_t),
+						ss->sec_hash_seed);
+	uint32_t mask = ~0;
+	uint32_t bit_loc;
+
+	for (j = 0; j < ss->num_hashes; j++) {
+		bit_loc = (h1 + j * h2) & ss->bit_mask;
+		mask &= test_bit(bit_loc, ss);
+	}
+
+	if (mask) {
+		*set_id = __builtin_ctzl(mask) + 1;
+		return 1;
+	}
+
+	*set_id = RTE_MEMBER_NO_MATCH;
+	return 0;
+}
+
+uint32_t
+rte_member_lookup_bulk_vbf(const struct rte_member_setsum *ss,
+		const void **keys, uint32_t num_keys, member_set_t *set_ids)
+{
+	uint32_t i, k;
+	uint32_t ret = 0;
+	uint32_t mask[RTE_MEMBER_LOOKUP_BULK_MAX];
+	uint32_t h1[RTE_MEMBER_LOOKUP_BULK_MAX], h2[RTE_MEMBER_LOOKUP_BULK_MAX];
+	uint32_t bit_loc;
+
+	for (i = 0; i < num_keys; i++)
+		h1[i] = MEMBER_HASH_FUNC(keys[i], ss->key_len,
+						ss->prim_hash_seed);
+	for (i = 0; i < num_keys; i++)
+		h2[i] = MEMBER_HASH_FUNC(&h1[i], sizeof(uint32_t),
+						ss->sec_hash_seed);
+	for (i = 0; i < num_keys; i++) {
+		mask[i] = ~0;
+		for (k = 0; k < ss->num_hashes; k++) {
+			bit_loc = (h1[i] + k * h2[i]) & ss->bit_mask;
+			mask[i] &= test_bit(bit_loc, ss);
+		}
+	}
+	for (i = 0; i < num_keys; i++) {
+		if (mask[i]) {
+			set_ids[i] = __builtin_ctzl(mask[i]) + 1;
+			ret++;
+		} else
+			set_ids[i] = RTE_MEMBER_NO_MATCH;
+	}
+	return ret;
+}
+
+uint32_t
+rte_member_lookup_multi_vbf(const struct rte_member_setsum *ss,
+		const void *key, uint32_t match_per_key,
+		member_set_t *set_id)
+{
+	uint32_t ret = 0;
+	uint32_t j;
+	uint32_t h1 = MEMBER_HASH_FUNC(key, ss->key_len, ss->prim_hash_seed);
+	uint32_t h2 = MEMBER_HASH_FUNC(&h1, sizeof(uint32_t),
+						ss->sec_hash_seed);
+	uint32_t mask = ~0;
+	uint32_t bit_loc;
+
+	for (j = 0; j < ss->num_hashes; j++) {
+		bit_loc = (h1 + j * h2) & ss->bit_mask;
+		mask &= test_bit(bit_loc, ss);
+	}
+	while (mask) {
+		uint32_t loc = __builtin_ctzl(mask);
+		set_id[ret] = loc + 1;
+		ret++;
+		if (ret >= match_per_key)
+			return ret;
+		mask &= ~(1U << loc);
+	}
+	return ret;
+}
+
+uint32_t
+rte_member_lookup_multi_bulk_vbf(const struct rte_member_setsum *ss,
+		const void **keys, uint32_t num_keys, uint32_t match_per_key,
+		uint32_t *match_count,
+		member_set_t *set_ids)
+{
+	uint32_t i, k;
+	uint32_t ret = 0;
+	uint32_t match_cnt_t;
+	uint32_t mask[RTE_MEMBER_LOOKUP_BULK_MAX];
+	uint32_t h1[RTE_MEMBER_LOOKUP_BULK_MAX], h2[RTE_MEMBER_LOOKUP_BULK_MAX];
+	uint32_t bit_loc;
+
+	for (i = 0; i < num_keys; i++)
+		h1[i] = MEMBER_HASH_FUNC(keys[i], ss->key_len,
+						ss->prim_hash_seed);
+	for (i = 0; i < num_keys; i++)
+		h2[i] = MEMBER_HASH_FUNC(&h1[i], sizeof(uint32_t),
+						ss->sec_hash_seed);
+	for (i = 0; i < num_keys; i++) {
+		mask[i] = ~0;
+		for (k = 0; k < ss->num_hashes; k++) {
+			bit_loc = (h1[i] + k * h2[i]) & ss->bit_mask;
+			mask[i] &= test_bit(bit_loc, ss);
+		}
+	}
+	for (i = 0; i < num_keys; i++) {
+		match_cnt_t = 0;
+		while (mask[i]) {
+			uint32_t loc = __builtin_ctzl(mask[i]);
+			set_ids[i * match_per_key + match_cnt_t] = loc + 1;
+			match_cnt_t++;
+			if (match_cnt_t >= match_per_key)
+				break;
+			mask[i] &= ~(1U << loc);
+		}
+		match_count[i] = match_cnt_t;
+		if (match_cnt_t != 0)
+			ret++;
+	}
+	return ret;
+}
+
+int
+rte_member_add_vbf(const struct rte_member_setsum *ss,
+		const void *key, member_set_t set_id)
+{
+	uint32_t i, h1, h2;
+	uint32_t bit_loc;
+
+	if (set_id > ss->num_set || set_id == RTE_MEMBER_NO_MATCH)
+		return -EINVAL;
+
+	h1 = MEMBER_HASH_FUNC(key, ss->key_len, ss->prim_hash_seed);
+	h2 = MEMBER_HASH_FUNC(&h1, sizeof(uint32_t), ss->sec_hash_seed);
+
+	for (i = 0; i < ss->num_hashes; i++) {
+		bit_loc = (h1 + i * h2) & ss->bit_mask;
+		set_bit(bit_loc, ss, set_id);
+	}
+	return 0;
+}
+
+void
+rte_member_free_vbf(struct rte_member_setsum *ss)
+{
+	rte_free(ss->table);
+}
+
+void
+rte_member_reset_vbf(const struct rte_member_setsum *ss)
+{
+	uint32_t *vbf = ss->table;
+	memset(vbf, 0, (ss->num_set * ss->bits) >> 3);
+}
diff --git a/lib/librte_member/rte_member_vbf.h b/lib/librte_member/rte_member_vbf.h
new file mode 100644
index 0000000..16144ff
--- /dev/null
+++ b/lib/librte_member/rte_member_vbf.h
@@ -0,0 +1,79 @@ 
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2017 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * 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.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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 COPYRIGHT
+ *   OWNER 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.
+ */
+
+#ifndef _RTE_MEMBER_VBF_H_
+#define _RTE_MEMBER_VBF_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+int
+rte_member_create_vbf(struct rte_member_setsum *ss,
+		const struct rte_member_parameters *params);
+
+int
+rte_member_lookup_vbf(const struct rte_member_setsum *setsum,
+		const void *key, member_set_t *set_id);
+
+uint32_t
+rte_member_lookup_bulk_vbf(const struct rte_member_setsum *setsum,
+		const void **keys, uint32_t num_keys,
+		member_set_t *set_ids);
+
+uint32_t
+rte_member_lookup_multi_vbf(const struct rte_member_setsum *setsum,
+		const void *key, uint32_t match_per_key,
+		member_set_t *set_id);
+
+uint32_t
+rte_member_lookup_multi_bulk_vbf(const struct rte_member_setsum *setsum,
+		const void **keys, uint32_t num_keys, uint32_t match_per_key,
+		uint32_t *match_count,
+		member_set_t *set_ids);
+
+int
+rte_member_add_vbf(const struct rte_member_setsum *setsum,
+		const void *key, member_set_t set_id);
+
+void
+rte_member_free_vbf(struct rte_member_setsum *ss);
+
+void
+rte_member_reset_vbf(const struct rte_member_setsum *setsum);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_MEMBER_VBF_H_ */