Let's talk about the elimination strategy of Redis cache

Redis What are the elimination strategies for cache? This article will talk with you about the Redis cache elimination strategy and introduce the cache strategy setting suggestions. I hope it will be helpful to you!

Redis (Remote Dictionary Server), the remote dictionary service, is an open source log type written in ANSI C language, supports the network, and can be based on memory or persistence. , Key-Value database, and provide APIs in multiple languages. [Related recommendations: Redis Video Tutorial]

It has the following characteristics:

• It runs based on memory and has high performance features
• Supports distributed, theoretically unlimited expansion of the
• key-value storage structure, efficient query
• provides multiple development language APIs, and is easy to integrate with existing business systems.

Usually used in business systems for distributed cache, centralized session storage, distributed locks and other application scenarios.

Whether it is local cache or distributed cache, in order to ensure higher performance, memory is used to save data. Due to cost and memory limitations, when the stored data exceeds the cache capacity, the cached data needs to be cached. Cull. Common elimination strategies include FIFO to eliminate the oldest data, LRU to eliminate the least recently used data, and LFU to eliminate the least recently used data.

Redis cache elimination policy trigger

In the production environment, we do not allow redis to have swap behavior. Therefore, the maximum memory usage is generally limited. Redis provides the configuration parameter maxmemory to specify the maximum memory usage.

The following configurations are legal:

maxmemory 1000KB 
maxmemory 100MB 
maxmemory 1GB 
maxmemory 0  # 表示不做限制，一般不会用

redis.conf configuration file is as follows

##8 8 Redis cache strategies

• volatile-lru deletes the least commonly used data among the data with a set timeout;

• allkeys-lru queries all keys Delete the least frequently used data in the data, which is the most widely used strategy;

volatile-random randomly deletes the data that has set a timeout;
allkeys- random queries all keys and then deletes them randomly;
volatile-ttl queries all data with a set timeout, sorts them immediately, and deletes data that is about to expire;
noeviction ( Default) If set to this attribute, no deletion operation will be performed, and an error will be returned if the memory overflows;
• volatile-lfu evicts the least frequently used key from all keys configured with an expiration time. ;

allkeys-lfu evicts the least frequently used keys from all keys;

Redis-based LRU and LFU algorithms

LRU algorithm

The Redis LRU algorithm is not an exact implementation. This means that Redis cannot select the

besteviction candidate, i.e. the one with the most visits in the past. Instead, it attempts to run an approximation of the LRU algorithm by sampling a small number of keys and then evicting the best (with the earliest access time) of the sampled keys. However, starting with Redis 3.0, the algorithm has been improved and can also select some good candidates for eviction. This improves the performance of the algorithm, allowing it to more closely resemble the behavior of the real LRU algorithm.

The important thing about the Redis LRU algorithm is that you

can

tune the accuracy of the algorithm by changing the number of samples to check for each eviction. This parameter is controlled by the following configuration directive:

maxmemory-samples 5

The reason Redis does not use a true LRU implementation is because it requires more memory. However, for applications using Redis, the approximation is actually equivalent. The following is a comparison chart between the LRU approximation used by Redis and the real LRU.

#The test that generates the above chart populates a Redis server with a given number of keys. Keys are accessed from first to last, so the first key is the best candidate for eviction using the LRU algorithm. An additional 50% of the keys were later added to force eviction of half of the old keys.

You can see three kinds of points in the picture, forming three different bands.

The light gray band is the evicted object.

• The gray band is an object that has not been evicted.

The green band is the added object.

In a theoretical LRU implementation, we would expect that of the old keys, the first half will expire. The Redis LRU algorithm will only expire old keys with

probability

. LRU is simply a model that predicts the likelihood that a given key will be accessed in the future. Additionally, if your data access pattern closely resembles a power law, most accesses will be in key sets that the LRU approximation algorithm can handle well.

Disadvantages: A large amount of cold data may be accessed within a certain period of time to generate a large amount of hot data

LFU 算法

从 Redis 4.0 开始，可以使用新的最不常用驱逐模式。这种模式在某些情况下可能会更好（提供更好的命中率/未命中率），因为使用 LFU Redis 会尝试跟踪项目的访问频率，因此很少使用的项目会被驱逐，而经常使用的项目有更高的机会留在记忆中。

如果您认为在 LRU，最近访问过但实际上几乎从未被请求过的项目不会过期，因此风险是驱逐将来有更高机会被请求的密钥。LFU 没有这个问题，一般应该更好地适应不同的访问模式。

配置LFU模式，可以使用以下策略：

• volatile-lfu 在具有过期集的键中使用近似 LFU 驱逐。
• allkeys-lfu 使用近似 LFU 驱逐任何密钥。

LFU 类似于 LRU：它使用一个概率计数器，称为莫里斯计数器，以便仅使用每个对象的几位来估计对象访问频率，并结合衰减周期，以便计数器随着时间的推移而减少：在某些时候，我们不再希望将密钥视为经常访问的密钥，即使它们过去是这样，以便算法可以适应访问模式的转变。

这些信息的采样与 LRU 发生的情况类似（如本文档的前一部分所述），以便选择驱逐的候选人。

然而，与 LRU 不同的是，LFU 具有某些可调参数：例如，如果不再访问频繁项，它的排名应该以多快的速度降低？还可以调整 Morris 计数器范围，以便更好地使算法适应特定用例。

默认情况下，Redis 4.0 配置为：

• 在大约一百万个请求时使计数器饱和。
• 每一分钟衰减一次计数器。

这些应该是合理的值并经过实验测试，但用户可能希望使用这些配置设置以选择最佳值。

有关如何调整这些参数的说明可以redis.conf在源代码分发的示例文件中找到，但简单地说，它们是：

lfu-log-factor 10 
lfu-decay-time 1

衰减时间是显而易见的，它是计数器应该衰减的分钟数，当采样并发现它比该值更旧时。一个特殊值0意味着：每次扫描时总是衰减计数器，很少有用。

计数器对数因子会改变需要多少次命中才能使频率计数器饱和，这恰好在 0-255 的范围内。系数越高，需要越多的访问以达到最大值。根据下表，系数越低，低访问计数器的分辨率越好：

+--------+------------+------------+------------+------------+------------+
| factor | 100 hits   | 1000 hits  | 100K hits  | 1M hits    | 10M hits   |
+--------+------------+------------+------------+------------+------------+
| 0      | 104        | 255        | 255        | 255        | 255        |
+--------+------------+------------+------------+------------+------------+
| 1      | 18         | 49         | 255        | 255        | 255        |
+--------+------------+------------+------------+------------+------------+
| 10     | 10         | 18         | 142        | 255        | 255        |
+--------+------------+------------+------------+------------+------------+
| 100    | 8          | 11         | 49         | 143        | 255        |
+--------+------------+------------+------------+------------+------------+

淘汰最近一段时间被访问次数最少的数据，以次数作为参考。

缺点：

1. 最近加入的数据常常容易被剔除，因为其起始方法次数比较少，

2. 如果频率时间度量为 1 个小时，则平均一天每个小时内访问频率 1000 的热点数据可能会被 2个小时的一段时间访问的频率为 1001 的数据剔除掉。可能会出现一些临界值的数据。

缓存策略设置建议

建议：了解Redis 的淘汰策略之后，在平时使用尽量主动设置/更新 key 的 expire 时间主动剔除不活跃的旧数据， 有助于提升查询性能

更多编程相关知识，请访问：编程入门！！

The above is the detailed content of Let's talk about the elimination strategy of Redis cache. For more information, please follow other related articles on the PHP Chinese website!