2009-07-08 13:19:16 -07:00
|
|
|
/*
|
hash: Replace primary hash functions by murmurhash.
murmurhash is faster than Jenkins and slightly higher quality, so switch to
it for hashing words.
The best timings I got for hashing for data lengths of the following
numbers of 32-bit words, in seconds per 1,000,000,000 hashes, were:
words murmurhash Jenkins hash
----- ---------- ------------
1 8.4 10.4
2 10.3 10.3
3 11.2 10.7
4 12.6 18.0
5 13.9 18.3
6 15.2 18.7
In other words, murmurhash outperforms Jenkins for all input lengths other
than exactly 3 32-bit words (12 bytes). (It's understandable that Jenkins
would have a best case at 12 bytes, because Jenkins works in 12-byte
chunks.) Even in the case where Jenkins is faster, it's only by 5%. On
average within this data set, murmurhash is 15% faster, and for 4-word
input it is 30% faster.
We retain Jenkins for flow_hash_symmetric_l4() and flow_hash_fields(),
which are cases where the hash value is exposed externally.
This commit appears to improve "ovs-benchmark rate" results slightly by
a few hundred connections per second (under 1%), when used with an NVP
controller.
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
2013-01-16 16:14:42 -08:00
|
|
|
* Copyright (c) 2008, 2009, 2010, 2012, 2013 Nicira, Inc.
|
2009-07-08 13:19:16 -07:00
|
|
|
*
|
2009-06-15 15:11:30 -07:00
|
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
* you may not use this file except in compliance with the License.
|
|
|
|
* You may obtain a copy of the License at:
|
2009-07-08 13:19:16 -07:00
|
|
|
*
|
2009-06-15 15:11:30 -07:00
|
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
*
|
|
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
* See the License for the specific language governing permissions and
|
|
|
|
* limitations under the License.
|
2009-07-08 13:19:16 -07:00
|
|
|
*/
|
|
|
|
#include <config.h>
|
|
|
|
#include "hash.h"
|
|
|
|
#include <string.h>
|
2012-04-09 14:33:33 -07:00
|
|
|
#include "unaligned.h"
|
2009-07-08 13:19:16 -07:00
|
|
|
|
2010-12-16 13:53:29 -08:00
|
|
|
/* Returns the hash of 'a', 'b', and 'c'. */
|
2009-10-14 17:03:55 -07:00
|
|
|
uint32_t
|
2010-12-16 13:53:29 -08:00
|
|
|
hash_3words(uint32_t a, uint32_t b, uint32_t c)
|
2009-10-14 17:03:55 -07:00
|
|
|
{
|
2014-07-04 07:57:18 -07:00
|
|
|
return hash_finish(hash_add(hash_add(hash_add(a, 0), b), c), 12);
|
2010-12-16 13:53:29 -08:00
|
|
|
}
|
|
|
|
|
2009-07-08 13:19:16 -07:00
|
|
|
/* Returns the hash of the 'n' bytes at 'p', starting from 'basis'. */
|
|
|
|
uint32_t
|
|
|
|
hash_bytes(const void *p_, size_t n, uint32_t basis)
|
|
|
|
{
|
2013-07-22 15:47:19 -07:00
|
|
|
const uint32_t *p = p_;
|
hash: Replace primary hash functions by murmurhash.
murmurhash is faster than Jenkins and slightly higher quality, so switch to
it for hashing words.
The best timings I got for hashing for data lengths of the following
numbers of 32-bit words, in seconds per 1,000,000,000 hashes, were:
words murmurhash Jenkins hash
----- ---------- ------------
1 8.4 10.4
2 10.3 10.3
3 11.2 10.7
4 12.6 18.0
5 13.9 18.3
6 15.2 18.7
In other words, murmurhash outperforms Jenkins for all input lengths other
than exactly 3 32-bit words (12 bytes). (It's understandable that Jenkins
would have a best case at 12 bytes, because Jenkins works in 12-byte
chunks.) Even in the case where Jenkins is faster, it's only by 5%. On
average within this data set, murmurhash is 15% faster, and for 4-word
input it is 30% faster.
We retain Jenkins for flow_hash_symmetric_l4() and flow_hash_fields(),
which are cases where the hash value is exposed externally.
This commit appears to improve "ovs-benchmark rate" results slightly by
a few hundred connections per second (under 1%), when used with an NVP
controller.
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
2013-01-16 16:14:42 -08:00
|
|
|
size_t orig_n = n;
|
|
|
|
uint32_t hash;
|
2009-07-08 13:19:16 -07:00
|
|
|
|
hash: Replace primary hash functions by murmurhash.
murmurhash is faster than Jenkins and slightly higher quality, so switch to
it for hashing words.
The best timings I got for hashing for data lengths of the following
numbers of 32-bit words, in seconds per 1,000,000,000 hashes, were:
words murmurhash Jenkins hash
----- ---------- ------------
1 8.4 10.4
2 10.3 10.3
3 11.2 10.7
4 12.6 18.0
5 13.9 18.3
6 15.2 18.7
In other words, murmurhash outperforms Jenkins for all input lengths other
than exactly 3 32-bit words (12 bytes). (It's understandable that Jenkins
would have a best case at 12 bytes, because Jenkins works in 12-byte
chunks.) Even in the case where Jenkins is faster, it's only by 5%. On
average within this data set, murmurhash is 15% faster, and for 4-word
input it is 30% faster.
We retain Jenkins for flow_hash_symmetric_l4() and flow_hash_fields(),
which are cases where the hash value is exposed externally.
This commit appears to improve "ovs-benchmark rate" results slightly by
a few hundred connections per second (under 1%), when used with an NVP
controller.
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
2013-01-16 16:14:42 -08:00
|
|
|
hash = basis;
|
|
|
|
while (n >= 4) {
|
2014-07-04 07:57:18 -07:00
|
|
|
hash = hash_add(hash, get_unaligned_u32(p));
|
hash: Replace primary hash functions by murmurhash.
murmurhash is faster than Jenkins and slightly higher quality, so switch to
it for hashing words.
The best timings I got for hashing for data lengths of the following
numbers of 32-bit words, in seconds per 1,000,000,000 hashes, were:
words murmurhash Jenkins hash
----- ---------- ------------
1 8.4 10.4
2 10.3 10.3
3 11.2 10.7
4 12.6 18.0
5 13.9 18.3
6 15.2 18.7
In other words, murmurhash outperforms Jenkins for all input lengths other
than exactly 3 32-bit words (12 bytes). (It's understandable that Jenkins
would have a best case at 12 bytes, because Jenkins works in 12-byte
chunks.) Even in the case where Jenkins is faster, it's only by 5%. On
average within this data set, murmurhash is 15% faster, and for 4-word
input it is 30% faster.
We retain Jenkins for flow_hash_symmetric_l4() and flow_hash_fields(),
which are cases where the hash value is exposed externally.
This commit appears to improve "ovs-benchmark rate" results slightly by
a few hundred connections per second (under 1%), when used with an NVP
controller.
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
2013-01-16 16:14:42 -08:00
|
|
|
n -= 4;
|
2013-07-22 15:47:19 -07:00
|
|
|
p += 1;
|
2009-07-08 13:19:16 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
if (n) {
|
hash: Replace primary hash functions by murmurhash.
murmurhash is faster than Jenkins and slightly higher quality, so switch to
it for hashing words.
The best timings I got for hashing for data lengths of the following
numbers of 32-bit words, in seconds per 1,000,000,000 hashes, were:
words murmurhash Jenkins hash
----- ---------- ------------
1 8.4 10.4
2 10.3 10.3
3 11.2 10.7
4 12.6 18.0
5 13.9 18.3
6 15.2 18.7
In other words, murmurhash outperforms Jenkins for all input lengths other
than exactly 3 32-bit words (12 bytes). (It's understandable that Jenkins
would have a best case at 12 bytes, because Jenkins works in 12-byte
chunks.) Even in the case where Jenkins is faster, it's only by 5%. On
average within this data set, murmurhash is 15% faster, and for 4-word
input it is 30% faster.
We retain Jenkins for flow_hash_symmetric_l4() and flow_hash_fields(),
which are cases where the hash value is exposed externally.
This commit appears to improve "ovs-benchmark rate" results slightly by
a few hundred connections per second (under 1%), when used with an NVP
controller.
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
2013-01-16 16:14:42 -08:00
|
|
|
uint32_t tmp = 0;
|
2012-04-09 14:33:33 -07:00
|
|
|
|
hash: Replace primary hash functions by murmurhash.
murmurhash is faster than Jenkins and slightly higher quality, so switch to
it for hashing words.
The best timings I got for hashing for data lengths of the following
numbers of 32-bit words, in seconds per 1,000,000,000 hashes, were:
words murmurhash Jenkins hash
----- ---------- ------------
1 8.4 10.4
2 10.3 10.3
3 11.2 10.7
4 12.6 18.0
5 13.9 18.3
6 15.2 18.7
In other words, murmurhash outperforms Jenkins for all input lengths other
than exactly 3 32-bit words (12 bytes). (It's understandable that Jenkins
would have a best case at 12 bytes, because Jenkins works in 12-byte
chunks.) Even in the case where Jenkins is faster, it's only by 5%. On
average within this data set, murmurhash is 15% faster, and for 4-word
input it is 30% faster.
We retain Jenkins for flow_hash_symmetric_l4() and flow_hash_fields(),
which are cases where the hash value is exposed externally.
This commit appears to improve "ovs-benchmark rate" results slightly by
a few hundred connections per second (under 1%), when used with an NVP
controller.
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
2013-01-16 16:14:42 -08:00
|
|
|
memcpy(&tmp, p, n);
|
2014-07-04 07:57:18 -07:00
|
|
|
hash = hash_add(hash, tmp);
|
2009-07-08 13:19:16 -07:00
|
|
|
}
|
|
|
|
|
2014-07-04 07:57:18 -07:00
|
|
|
return hash_finish(hash, orig_n);
|
2009-07-08 13:19:16 -07:00
|
|
|
}
|
2012-08-21 14:26:23 -07:00
|
|
|
|
hash: Replace primary hash functions by murmurhash.
murmurhash is faster than Jenkins and slightly higher quality, so switch to
it for hashing words.
The best timings I got for hashing for data lengths of the following
numbers of 32-bit words, in seconds per 1,000,000,000 hashes, were:
words murmurhash Jenkins hash
----- ---------- ------------
1 8.4 10.4
2 10.3 10.3
3 11.2 10.7
4 12.6 18.0
5 13.9 18.3
6 15.2 18.7
In other words, murmurhash outperforms Jenkins for all input lengths other
than exactly 3 32-bit words (12 bytes). (It's understandable that Jenkins
would have a best case at 12 bytes, because Jenkins works in 12-byte
chunks.) Even in the case where Jenkins is faster, it's only by 5%. On
average within this data set, murmurhash is 15% faster, and for 4-word
input it is 30% faster.
We retain Jenkins for flow_hash_symmetric_l4() and flow_hash_fields(),
which are cases where the hash value is exposed externally.
This commit appears to improve "ovs-benchmark rate" results slightly by
a few hundred connections per second (under 1%), when used with an NVP
controller.
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Ethan Jackson <ethan@nicira.com>
2013-01-16 16:14:42 -08:00
|
|
|
uint32_t
|
|
|
|
hash_double(double x, uint32_t basis)
|
|
|
|
{
|
|
|
|
uint32_t value[2];
|
|
|
|
BUILD_ASSERT_DECL(sizeof x == sizeof value);
|
|
|
|
|
|
|
|
memcpy(value, &x, sizeof value);
|
|
|
|
return hash_3words(value[0], value[1], basis);
|
|
|
|
}
|
2014-07-11 05:57:11 -07:00
|
|
|
|
|
|
|
uint32_t
|
|
|
|
hash_words__(const uint32_t p[], size_t n_words, uint32_t basis)
|
|
|
|
{
|
|
|
|
return hash_words_inline(p, n_words, basis);
|
|
|
|
}
|
|
|
|
|
|
|
|
uint32_t
|
|
|
|
hash_words64__(const uint64_t p[], size_t n_words, uint64_t basis)
|
|
|
|
{
|
|
|
|
return hash_words64_inline(p, n_words, basis);
|
|
|
|
}
|