商用过滤器哪个牌子好_filter前置过滤器

(36) 2024-09-07 19:01:01

  • 我们称quotient filter商过滤器,与布鲁姆过滤器(Bloom filter)不同的是,商过滤器采用一个哈希函数,具有更高的查找效率,它将关键字key经过一次哈希运算后,得到q_hash然后将q_hash/2^r 作为商r=q_hash%2^r作为余数
  • 我们构造一个记录器,它由32r个bit组成,每3个bit对应个槽,那么总共就有2^r个槽,每个槽中存放余数r
  • 当插入一个元素的时候,先计算出商和余数,由商确定槽的位置,从而决定应该插入到哪一个槽中,但是这个槽完全有可能被其他元素占用,所以这个时候,应该比较余数的大小,如果占用的那个元素的余数比新加入进来的余数大,那么就将占用的余数向右移动一个槽位,同时把新加入的余数存放到该槽,如果新加入进来的余数比槽中的余数大,那么就加入到该槽后面合适的位置就可以了,简单的说,就是插入排序的思想。
  • 这样,就使得一个元素从初始槽开始,连续占用了批量的槽,我们称它为一个集群,它们有一个共同的特点,其他元素都是由于自己本身的槽被占用,而被迫放到后面的槽中。
  • 所以,这就意味着,当前槽有三种状态,与槽对应的三个bit相关联

is_occupied表示是否已经有商号对应该槽,如果是被迫转移过来的,那么不应该将其设置成1,只有在计算出某个商号对应该槽号后,才将其设置成1
is_continuation不是放在自己本身的槽中,被迫后移,与前面的元素形成连续集群
is_shifed 是不是放在与自己商号对应的槽中,如果不是,被迫后移动的,那么将该状态设置成1

  • 由上面的三种状态表明,状态的开始标志位is_coninuation如果是0,就表明一个集群的开始 当查询一个元素的时候,先将元素经过一次哈希运算得到q_hash,然后再得到商号和余数
  • 由商号可以得到在记录器中的下标,由于在插入元素时的特性,这就意味着,该元素对应的余数只会存放到该元素对应的商号以及后面的槽号,该元素的查询范围,从一个集群的开始,到该集群的结束,结束的标志为:一次往后探测发现一个空槽了,或者后面的is_continuation设置成0了,或者槽中的余数大于当前余数了,经过对比余数相等后,就证明找到了该元素,当然,这种查询方法也存在一定的假阳性误判率,如果哈哈希函数设计得当,那么这种查询是时间复杂度为O(1),当然,如果哈希函数设计不得当,这种查询的时间复杂度完全可能变成O(m),m为记录器的长度。
  • 商用过滤器哪个牌子好_filter前置过滤器 (https://mushiming.com/)  第1张
    下面再给出文章对于quotient filter(商过滤器)的讲解

Quotient Filter

  • The QF is a space-efficient and cache-friendly probabilistic data structure that uses the quotienting technique of hashing [12] to store a set S⊂U efficiently. Mapping is done for every element × ∈S to h(x), where h(x) is a primary hash function resulting in a set of p bits named as the fingerprint of x, that is, h(x)↦{0,…,2P−1}⇒fp(x).
  • fp(x) is an open hash table with m=2q buckets where each bucket has (r+3) bits. In fp(x), the least significant bits are denoted by r, and the most significant bits in quotienting are represented by q=(p−r). Insertion operation in the QF is performed by computing the quotient fq←(⌊fp(x)/2r⌋) and remainder fr←(fp(x)mod 2r) of every considered element. Here, the index of the bucket used for inserting an element is denoted by fq, whereas fr represents a value inserted in the bucket fq.
  • Two important terms used for identifying the appropriate positions for insertion and querying in QF are run and cluster. Run refers to a scenario where remainders of different fingerprints having the same fq are stored contiguously, that is, fq of two items collide but fr are distinct. Such collisions are resolved through linear probing. In such scenarios, remainders associated with different fq are shifted and corresponding meta-data bits are updated for each bucket if required. A cluster is a sequence of one or more consecutive runs with no empty bucket between them. A cluster is immediately followed by an empty slot.
  • A general observation is that a Bloom filter (BF) has more hashing functions as compared to a QF. Since hash functions are generated only once per QF and every signature is implemented using one location, single-memory access is required to check the presence of a signature. Thus, in comparison to BF, QF has higher throughput.

参考:

[1].博客
[2].Edge Computing-Based Security Framework for Big Data Analytics in VANETs

THE END

发表回复